# Create a Custom AI Agent Source: https://guides.robylon.ai/ai-agents/custom-ai-agents Build specialized agents to handle specific tasks within your workflows. Custom AI agents allow you to control behavior, assign tools, and fine-tune models for task-oriented automation. ## What is a Custom AI Agent? A **Custom AI Agent** is a task-focused assistant that: * Runs inside your workflows * Has its own **prompt, tools, and context** * Operates on a selected **channel** (e.g., chat, voice, or tickets) * Can be tuned for accuracy, creativity, or brevity depending on use case Typical use cases include: * **Order status agents** that fetch and explain shipping updates * **Billing agents** that answer invoice questions * **Returns agents** that handle policy queries with tool access ## Steps to Create a Custom AI Agent Go to **Workflows → Settings → AI Agents**. Click **Add Agent** to start the creation process. * **Name** — Give your agent a clear name * **Purpose** — Add a short description (e.g., “Order Support Agent”) * **Channel** — Select where it operates (e.g., Chat) * **Prompt & Context** — Provide a detailed system prompt and background info * **Add Tools** — Assign APIs or functions the agent can call * **User Info** — Pass variables (e.g., `order_id`, `email`) for customer context * **Fine-Tune Model** — Adjust model type, temperature, and max tokens Click **Save** to store your custom configuration. Add a **Custom AI Agent block** in any workflow to execute its function. # Configure the Knowledge Base (KB) Agent Source: https://guides.robylon.ai/ai-agents/kb-agent A Knowledge Base (KB) Agent is created automatically when you add documents to your knowledge base. It leverages these documents to answer user questions and can be enhanced with instructions, tools, and personalization. ## Provide Custom Instructions In the **KB agent panel**, add **custom instructions** to guide how the agent queries information from your documents.\ Examples: * “Always cite document titles when giving answers.” * “For shipping policy questions, reference only the official policy doc.” Use clear, rule-based instructions to keep answers aligned with company policies. ## Provide Tools for Dynamic Data The KB agent can also use **tools** for real-time or dynamic data. * Assign APIs or existing tools that supplement static documents. * Example: Provide a **subscription status API** so the agent checks a user’s account before answering. This enables **context-aware answers** that combine static documentation with live data. ## Personalize Responses with Variables Pass **user variables** into the KB agent for precise, personalized responses. * Example: Pass `user_type = premium`. * The agent can then provide **14-day returns** for premium users and **7-day returns** for standard users. Variables allow one KB agent to serve multiple customer tiers without duplicating content. ## Adjust AI Settings Fine-tune the KB agent’s performance in the **AI settings**: * **Model** — choose the underlying AI model (e.g., GPT-4, GPT-4o) * **Temperature** — lower for consistent support answers, higher for creativity * **Max tokens** — control the length of responses Start with **low temperature (0.2–0.4)** for accuracy. Increase only if responses feel too rigid. *** # Create a Triage (Master) AI Agent Source: https://guides.robylon.ai/ai-agents/triage-agent A Triage Agent (aka Master Agent) acts as a smart router. It analyzes each incoming ticket/chat, applies priority-based rules, and hands off to the right Sub-Agent for expert handling—rather than solving the issue itself. ## Key Concepts ### **Triage Agent (Master Agent)** A high-level orchestrator that reads incoming data, follows rule-based **Custom Instructions**, and **routes** the task to a Sub-Agent. ### **Sub-Agents** Specialized agents for domains such as **Billing**, **Technical Support**, **Account Management**, or **Sales**. Each owns a narrow scope and clear success criteria. ### **Custom Instructions** The agent’s “playbook.” A detailed, step-by-step prompt that defines **priorities**, **checks**, and **handover rules**. ### **Tools** External functions/APIs the agent can call to make better routing decisions (e.g., **check country**, **lookup order**, **subscription status**). *** ## How the Triage Process Works 1. **Initial Analysis** — Parse subject, description, user data, tags, and variables. 2. **Priority-Based Checks** — Execute checks in order: * **Priority 0: Pre-qualification & Filtering** (e.g., non-operational country → send policy message & close). * **Priority 1: Intent Identification** (e.g., `password_reset`, `billing_dispute`, `product_feature_question`). * **Priority 2: Keyword/Tag Fallback** when intent confidence is low. 3. **Routing** — Map **intent** (or fallback outcome) to a **Sub-Agent** and **handover**. 4. **Exception/Fallback** — If nothing matches, **route to General Queries** or **flag for human review**. flowchart TD A\[Incoming request] --> B\[Priority 0: Pre-checks] B -->|Non-operational country| X\[Auto-response / Close] B -->|OK| C\[Priority 1: Detect intent] C -->|High confidence| D\[Map to sub-agent] C -->|Low/None| E\[Priority 2: Keyword/Tag fallback] E -->|Found| D E -->|Not found| F\[Fallback: General agent or human review] D --> G\[Handover to Sub-Agent] *** ## Step-by-Step: Create the Triage Agent Ensure specialized agents exist (e.g., **Order Status**, **Withdrawals**, **Refund**). Each should have a clear scope, inputs, and tools. Navigate to **AI Agents** → click **Add Agent** → name it (e.g., **Master Triage Agent**) → select **Create Agent**. Add lookups you’ll reference in rules (e.g., **country check**, **order lookup**, **subscription status**). Provide a **role**, **process**, **priority rules**, **routing map**, and a **final fallback**. Save, test with sandbox cases, then publish to production when stable. *** # Create and Manage AI Personas Source: https://guides.robylon.ai/ai-persona An AI Persona defines the framing, tone, and style of replies sent to your customers. You can configure multiple personas to adapt communication for different channels, use cases, or teams. ## Quickstart — Create Your First Persona Go to **Settings → Team → AI Personas**.\ This is where you will find and manage your personas. Click the **Add AI Persona** button to begin the creation process. Fill in the following fields: * **Name** — A clear, identifiable name. * **Channels** — Select where the persona will apply (e.g., Email, Chat). * **Bio** — A detailed description that shapes AI responses. * **Tonality** — Choose tone: *Friendly*, *Casual*, *Professional*, or *Direct*. * **Language** — Primary and additional reply languages. * **Custom Instructions** — Special instructions for unique scenarios. Use the **Set as Default** option to bind this persona to a specific channel.\ All replies from that channel will then use this persona automatically. Click **Save** to store and activate the persona configuration. # Guide: Using the Analytics Dashboard Source: https://guides.robylon.ai/analytics The Analytics tab provides detailed reports and real-time data on team performance and customer interactions. This guide explains how to configure your dashboard, set up automated reports, and track key support metrics. ## Configure Your Dashboard Use the **configuration button** on the Analytics tab to enable or disable the analytics modules you want to view.\ You can personalize the dashboard for different roles (e.g., managers, team leads, agents). *** ## Create an Automated Report Click the **New Report** button on the Analytics tab. Choose how often the report is emailed: * **Daily** — Track day-to-day performance * **Monthly** — Monitor trends and strategy outcomes * **On-demand** — Generate custom reports anytime Export reports to an **Excel file** with detailed ticket resolutions and automation data. *** ## View Key Metrics View a list of team members with: * Online / Offline status * Login and logout times * Total time available for chat Track: * Chats created * Chats closed by the bot * Chats closed by agents Filter by **Today, Yesterday, or Last 7 Days**, and compare with previous periods. Monitor overall **CSAT scores** across all channels, as well as: * Bot CSAT * Agent CSAT View **trends over time** to identify performance peaks and dips. Analyze the distribution of **common issues** across chats and tickets.\ Helps identify recurring customer pain points and prioritize fixes. Measure knowledge base effectiveness: * Total questions asked * Successfully answered by KB * Unanswered queries Use these insights to improve and expand your documentation. *** ## Voice Analytics Track how your **voice agent** performs across pickups, duration, and user engagement. Use the **date filters** (15 Days, 7 Days, Today) to switch time ranges and compare trends. ### Voice Resolution (KPIs) At the top of the page, the Voice Resolution strip shows today’s headline metrics (with % deltas vs. previous period): * **Voice Calls Triggered** — Total outbound/inbound voice tasks initiated. * **Voice Calls Picked Up** — Calls answered by users. * **Voice Calls Picked Up in First Try** — First-attempt connections (quality of list/routing). * **Call Tries per Pickup** — Average attempts needed to get one pickup (lower is better). * **Average Call Duration per Picked-Up Call** — Typical handle time for successful calls. * **Average Number of User Turns per Picked-Up Call** — Conversation depth/engagement signal. If **Call Tries per Pickup** rises or **First-Try Pickups** drop, review dialer windows, number reputation, and list quality. ### Call Analytics (Timeline) A time-series chart of call outcomes for the selected range: * **Calls triggered** (red) * **Calls picked up** (blue) * **Picked up first try** (green) Use this to spot **spikes** in calling or **pickup windows** (hours where users answer most). ### Retries per Pickup Line chart of **average retries before a successful pickup** across the day. * Sudden peaks suggest list quality issues or suboptimal call windows. * Sustained high values warrant **cooldown** rules or **time-of-day** adjustments. ### Call Duration (Bucketed) Histogram of **picked-up call durations** (e.g., `<10s`, `10–30s`, `30–120s`, `120–300s`, `300–600s`, `>600s`). * More **30–120s** bars usually indicate efficient verification/short workflows. * Many **less than 10s** calls may imply accidental pickups or poor intros. ### User Turns (Distribution) Bar chart of **user turns per picked-up call** (e.g., `0`, `1–3`, `3–6`, …). * **0 turns** after pickup → user hang-ups or IVR misfit. * **1–3 turns** suggests short confirmations; **6+** may need script simplification. *** ### Recommended Actions * **Improve pickup**: Test alternate **calling windows**, verify **CLID reputation**, and throttle retries to reduce **Call Tries per Pickup**. * **Tighten intro**: If \*\* less than 10s durations\*\* spike, shorten the greeting and clearly state **purpose + opt-out**. * **Guide to resolution**: If **User Turns = 0** or **1** dominate, add a quick **DTMF** or **yes/no** prompt to capture intent before hang-up. * **Policy & cadence**: Cap total attempts per user/day and add **cooldowns** after consecutive failures. All charts respect the **15 Days / 7 Days / Today** filters. Use the same range when comparing KPIs to timeline patterns. *** ## Best Practices Schedule **monthly automated reports** for leadership review, and **daily reports** for frontline managers. Use **Knowledge Base Gap reports** to prioritize new help articles. A 10% reduction in unanswered queries can significantly boost CSAT. # Brand Info Source: https://guides.robylon.ai/brand-info The **Brand Info** page allows you to configure your organization's brand details in Robylon. It enables you to add company information, brand details, website, and brand icon, as well as set the organization's timezone used across the system. **Accessing the Brand Info Page** Follow these steps to configure your brand details. 1. Log in to your Robylon dashboard. 2. From the left sidebar, click **Settings**. 3. On the Settings page, under General, select **Brand Info**. After clicking **Brand Info**, enter the details. # Brand Info Fields ### **1. Company Legal Name** Enter the official registered name of your company. This information is used only for billing purposes and will not be displayed to customers. ### **2. Brand Name** Enter the name of your brand as it should appear to customers. This name will be used in chatbot interactions. ### **3. Brand ID** The Brand ID is a unique identifier automatically generated by Robylon for your brand. This field is system-generated and cannot be modified manually. ### **4. Brand Description** Provide a short description of your business or product. This helps AI agents understand your brand context and respond more accurately to user queries. ### **5. Website** Enter the official website URL of your brand. This allows your AI agents to reference your organization's official web presence. ### **6. Brand Icon** Upload your brand logo or icon. Recommended format: * Square image * 72px × 72px size * Colored background works best The icon will appear in chatbot interfaces and other customer-facing interactions. ### **7. Timezone** Select your **organization's timezone** from the dropdown. This timezone will be applied across the organization and used for system activities, including timestamp on conversations and analytics. Once you click **Save**, a confirmation window appears asking whether you want to update the timezone for all users in the account. If you want all current users in the organization to follow the organization timezone, click **Yes**. # Channels Source: https://guides.robylon.ai/channels-connectors/channels The Channels tab allows you to manage all the external applications connected to the Robylon platform. From here, you can review existing integrations or connect new apps. ## Connected Apps Under **Connected Apps**, you can view and control integrations with: * **Ticketing systems** (e.g., Zendesk, Freshdesk) * **Messaging platforms** (e.g., WhatsApp, Slack) * **Productivity tools** (e.g., Google Sheets) * **CRMs and sales platforms** (e.g., HubSpot) You can enable/disable integrations or update their configuration as needed. *** ## Available Applications Beyond the currently connected apps, you can also browse a list of all supported applications, including: * **Slack** * **Zendesk** * **HubSpot** * Other common support, CRM, and communication platforms *** ## Connecting a New Application Navigate to the **Channels tab**. Choose the app you want to connect (e.g., Slack, HubSpot). Click the **Connect** button to start the integration process. Enter the required details (e.g., API key, credentials) and authorize permissions. Once connected, the app will appear under **Connected Apps**. *** ## Best Practices * **Use secure credentials** — always store API keys in **Secrets**. * **Review permissions** before authorizing integrations. * **Test connections** before deploying automations that depend on them. * **Audit Connected Apps** periodically to remove unused or outdated integrations. *** # Chatbot Settings Source: https://guides.robylon.ai/channels-connectors/chatbot-settings The Chatbot Settings tab gives you full control over the look, feel, and behavior of your chatbot. It also provides installation instructions for deploying on both web and mobile platforms. A live preview widget is displayed on the left to show how the chatbot will appear once deployed. ## Interface Configure how the chatbot appears on your website or app: * **Display modes**: * **Widget** (floating chat bubble) * **Popover** (overlay) * **Embed** (inline section) * **Positioning**: Control widget alignment, side spacing, and bottom spacing. * **Toggles**: * **Stream Text** — Visually stream text as it’s generated. * **Voice Input** — Allow users to speak via speech-to-text. * **Voice Output** — Use text-to-speech so the bot can talk. *** ## Styles Personalize the chatbot to match your brand: * **Text & Labels**: Set chatbot **title** and **subtitle**. * **Colors**: Choose **brand color** and **title-bar color**. * **Images**: * Widget icon * Agent avatar * Header image * **Banner & Text**: * Show a **welcome banner** with image and text * Customize **footer text** * Set a **message box placeholder** * Optionally include a **“Powered by Robylon”** link *** ## General Settings Adjust chatbot behaviors: * **Play sound** when a new message arrives * **Show AI indicator** for AI-generated replies * Allow users to **rate messages** *** ## Content Define chatbot display logic and conversation memory: * **Custom conditions** — control when the chatbot appears * **Delay** — set a delay before chatbot pops up * **Conversation history** — enable to let users view past chats When **conversation history** is enabled, you can also configure additional visual and branding elements for the chatbot experience, such as: * **Chatbot home page banner**\ You can set a welcoming banner that appears when users open the chatbot home page. This banner can display a custom image and message, creating a branded experience and helping guide users as they revisit previous conversations. * **Logo, heading, and subheading for the home page**\ You can add your company logo and define a heading with a short subheading to reinforce your brand identity. The heading introduces the chatbot’s purpose, while the subheading provides helpful context or instructions, making the home page look professional and user-friendly. ### Chatbot Home Page Banner * Add a **banner image** that appears at the top of the chatbot’s home screen. * Use this to highlight **brand identity**, share **seasonal campaigns**, or provide a **welcome graphic**. * Recommended size: ensure it is optimized for both desktop and mobile screens. ### Logo, Heading, and Subheading * **Logo** — Place your company or product logo prominently to reinforce trust and brand recognition. * **Heading** — Add a short, clear title (e.g., *Welcome to Robylon Support*). * **Subheading** — Provide a short description or guidance (e.g., *Ask me about your orders, subscriptions, or policies*). Together, these elements help frame the chatbot as a **branded, helpful entry point**—especially important when users are reviewing past conversations or starting new ones. *** ## Installation Find integration instructions for deploying the chatbot. ### Web Applications Supports installation on: * **JavaScript** * **React** * **Webflow** * **WordPress** * **Bubble** * **Framer** * **Wix** * **Shopify** ### Mobile Applications Supports installation on: * **Android** * **iOS** * **React Native** Follow the framework-specific instructions provided to ensure smooth setup and compatibility. *** # Voice Settings Source: https://guides.robylon.ai/channels-connectors/voice-settings The Voice Settings tab allows you to manage the phone numbers your voice agents use. You can view existing numbers, check providers, and add new ones for inbound or outbound calls. ## Active Phone Numbers * Displays all **active phone numbers** linked to your voice agents. * Shows the **provider** for each number (e.g., Twilio, Exotel). * Lets you edit or remove existing numbers as needed. *** ## Add a New Phone Number Open the **Voice Settings tab** and click **Add New Phone Number**. Choose between **Twilio** or **Exotel** as your provider. Provide the required information for the selected provider. Once saved, the number will appear in the list of active voice numbers. *** ## Twilio Configuration When adding a Twilio number, you must provide: * **Label** — Name to identify the number * **Phone Number** — The actual Twilio number * **Account SID** — Twilio account identifier * **Auth Token** — Secure token for authentication *** ## Exotel Configuration When adding an Exotel number, you must provide: * **Label** — Name to identify the number * **Phone Number** — The actual Exotel number * **Domain App ID** — Exotel app/domain identifier * **Account SID** — Exotel account identifier * **API Key** — Key for API authentication * **API Token** — Secure token for API authentication *** ## Best Practices * Store provider credentials securely in **Secrets**. * Use clear **labels** for numbers (e.g., *Support Line - India*). * Test new numbers with a **sample workflow** before going live. * Audit voice numbers periodically to remove unused or inactive ones. *** # Conversation Insights Source: https://guides.robylon.ai/conversation-insights The Conversation Insights tab helps you measure and analyze user sentiment in chats. By tracking sentiment at the message level, you can identify pain points, improve workflows, and guide agents with actionable recommendations. ## Key Metrics * **Net Sentiment Score (NSS)**\ The average sentiment across a conversation or set of conversations. * **Sentiment Shift Score (SSS)**\ Measures how sentiment changes throughout the chat (e.g., going from frustrated → reassured). * **Lowest Sentiment Point**\ Highlights the most negative user message with its sentiment score and detected emotion. * **Actionable Recommendations**\ AI-powered suggestions for next steps to resolve issues and improve customer experience. *** ## Sentiment Scoring Guide Each **user message** is scored between **-5 and +5**, based on the emotion expressed. | Score | Canonical Emotion Label | Umbrella Meaning | | ------ | ----------------------- | -------------------------------------------- | | **+5** | Ecstatic | Peak positivity — delight, gratitude, or joy | | **+4** | Delighted | Very positive satisfaction and excitement | | **+3** | Pleased | Moderately happy, content, optimistic | | **+2** | Satisfied | Mild approval or relief; things are on track | | **+1** | Reassured | Slight positive calm after concern | | **0** | Neutral | No strong emotion; observational | | **-1** | Uncertain | Mild doubt, hesitation, or confusion | | **-2** | Concerned | Noticeable worry or disappointment | | **-3** | Frustrated | Clear irritation or dissatisfaction | | **-4** | Angry | Strong negative feeling, indignation | | **-5** | Furious | Extreme negativity — outrage or despair | Track both **NSS** (overall mood) and **SSS** (direction of mood) to understand not just **how customers feel**, but **whether their experience improved or worsened** during the conversation. *** ## Use Cases * Identify **at-risk customers** with consistently low sentiment. * Review **Lowest Sentiment Points** to detect recurring issues. * Use **Actionable Recommendations** to guide agent training and automation improvements. * Track sentiment trends across **teams, workflows, and time periods** to measure impact of changes. *** # Customize your Robylon Bot Source: https://guides.robylon.ai/customize-your-robylon-bot Step-by-step to tailor tone, intents, guardrails, knowledge sources, and handoffs-so your AI agent matches your brand and resolves faster. **Step 1**: Log in to your Robylon account and open the dashboard. **Step 2**: In the left panel, click on the Settings button. Robylon dashboard with the left sidebar open and Settings highlighted to customize your bot **Step 3**: Within the Settings tab, select Channels. Settings → Channels screen in the Robylon dashboard for bot configuration **Step 4**: Click on Chatbot to open its interface. Channels view with Chatbot selected to open the bot customization interface **Step 5**: Now, here in the ‘Interface’ section, you can customize the chat interface and positioning of your chatbot. Chatbot Interface settings showing chat window layout and on-page position options **Step 6**: Then, click on the ‘Styles’ section, where you customize your chatbot’s appearance. * In this section, you can edit your chatbot title, subtitle, and colors of your chatbot. Chatbot Styles panel to edit bot title, subtitle, and brand colors * Scroll down in this section to customize the image assets of your bot, like the launcher icon, the header image, and the agent image of your bot.  Styles → Image assets for the bot: launcher icon, header image, and agent profile * Then, scroll down in this section to customize the footer of your chatbot, such as the message box placeholder. You can also add custom footer links in the footer. Styles → Footer customization for message placeholder text and custom footer links **Step 7**: Then, click on the ‘General’ section, where you can add an AI Disclaimer and User feedback in your chatbot, also you can add a message sound to your chatbot in this section. General settings to add an AI disclaimer, collect user feedback, and enable message sound **Step 8**: After changing every setting in each tab: Interface, Styles, and General, click on the update button in each tab to save your changes. Save changes using the Update button in the Interface, Styles, and General tabs # Create & Use APIs, Tools, and Functions Source: https://guides.robylon.ai/data-and-integrations/api-tools-functions Learn what APIs, Tools, and Functions are, when to use them, and how to implement them effectively with reusable templates, contracts, and Python code. Use **APIs**, **Tools**, and **Functions** for advanced customization and integrations. * **APIs** connect external systems and fetch/update data. * **Tools** package repeatable actions behind clean input/output contracts. * **Functions** run **Python** for custom logic that blocks can’t express. *** ## APIs ### What is an API? An **API template** defines how to call an external service (method, URL, headers, params/body) and where to store the response for use in workflows. **Best practices** * Centralize repeated calls as **saved templates**. * Use **secrets** for authentication (`${secrets.*}`). * Reference workflow variables (`${vars.*}`) for dynamic values. * Normalize responses and save them into `vars.*`. *** ### How to Create an API Template Go to the **APIs** section in the sidebar. Click **Add API** to create a new template. Define: * **Name** — e.g., `get_order_status` * **Request** — method (**GET**, **POST**) and URL * **Headers** — use secrets for auth tokens * **Params / Body** — JSON or key-value format * **Response Mapping** — select a variable to store results Click **Save**. The API is now reusable across workflows. ## Tools ### What is a Tool? A Tool is a reusable action that can be called by workflows or AI agents. Tools are best when multiple flows need the same logic with consistent inputs and outputs. ### Best practices * Convert tested workflow logic with Pre-convert Tool. * Keep tools stateless and idempotent. * Define clear input/output contracts. * Add a consistent error shape (code, message). ### How to Create a Tool * Design logic in a workflow. * Use Pre-convert Tool to save it as a reusable action. * The Tool appears in the Tools section with inputs and outputs inferred. ## Functions ### What is a Function? A Function is a block of Python code used when other workflow blocks aren’t flexible enough. Functions can validate data, run calculations, or transform values. ### Best practices * Keep functions small, pure, and testable. * Validate inputs early. * Use them for transforms, calculations, and glue code. * Avoid long scripts or network I/O (better suited to APIs/Tools). ## Testing & Troubleshooting * Dry-run APIs with sample variables; inspect responses. * Add internal logs for APIs, Tools, and Functions. * Use standard error shapes for consistency. * Keep functions lightweight; make tools reusable. * Prefer idempotent operations for safe retries. # Manage Secrets and API Keys Source: https://guides.robylon.ai/data-and-integrations/secrets-keys Use Secrets to securely store sensitive credentials and API Keys to give external systems controlled access to the Robylon API. ## Secrets ### What is a Secret? A **Secret** is a securely stored credential, such as a password, API token, or encryption key. Secrets can be referenced inside workflows without exposing their values. **Best practices** * Always use **secrets** instead of hardcoding values. * Limit access to administrators where possible. * Rotate secrets periodically for security compliance. * Use descriptive names (e.g., `zendesk_api_token`) for clarity. *** ### How to Create a Secret Go to the **Secrets** section in the sidebar. Click **Create Secret**. * **Name** — descriptive label for the secret * **Type** — choose the kind of secret (e.g., API key, password) * **Value** — paste the sensitive credential Use the access slider to restrict or allow **administrator visibility**.\ Once saved, the secret can be used in workflows without exposing the value. Secrets are **encrypted and hidden**. You cannot recover the raw value once stored — only update or replace it. *** ## API Keys ### What is an API Key? An API Key is a secure token that authorizes external systems or developers to call Robylon APIs. Each key is tied to your Robylon instance. These APIs can also be used to trigger voice calls and outbound voice messages. **Best practices** * Generate unique keys per integration or developer. * Revoke keys immediately if compromised. * Monitor usage via API logs. *** ### How to Create a Robylon API Key Go to the **API Keys** section. Click **Create API Key**. Provide this key to trusted external systems or developers to authenticate requests to your Robylon instance. Treat API keys like passwords. Never embed them in client-side code or public repositories. *** # Create Variables, Intents, and Entities Source: https://guides.robylon.ai/data-and-integrations/variables-intents-entities Learn how to define and configure variables, intents, and entities as core data objects for building automations. **Variables, Intents, and Entities** are the core data objects that power automations. * **Variables** store values used across workflows and agents. * **Intents** classify what a user is trying to do. * **Entities** extract structured data like IDs or numbers from unstructured input. ## Variables ### What is a Variable? A **Variable** is a data placeholder that holds a value, such as a string, number, or dynamic expression. Variables can be used globally across workflows, agents, and tools to maintain context. **Best practices** * Use **global variables** for values that multiple workflows share (e.g., company name, support email). * Use **dynamic expressions** to capture context from user messages. * Name variables clearly (e.g., `user_email`, `order_id`). *** ### How to Create a Global Variable Go to **Data & Integrations → Variables**. Click **Add Variable** and give it a descriptive name. Choose between: * **Static value** — e.g., “[support@company.com](mailto:support@company.com)” * **Dynamic expression** — value captured at runtime *** ## Intents ### What is an Intent? An **Intent** represents the purpose of a user’s message — e.g., *check order status*, *request refund*, *update email*. Intents help classify and route conversations. **Best practices** * Write clear intent names and descriptions. * Add **diverse utterances** that cover how users phrase requests. * Provide **positive and negative examples** of real tickets to refine accuracy. *** ### How to Create an Intent Go to the **Intents** section. Provide: * **Name** — e.g., “Refund Request” * **Description** — what this intent should classify Add user phrases (e.g., “I want my money back”, “Process my refund”). Provide: * **Positive examples** — tickets that belong to this intent * **Negative examples** — tickets that should not be classified under it *** ## Entities ### What is an Entity? An **Entity** is a structured piece of data extracted from user input. For example, an **Order ID**, **Transaction Number**, or **UTR Number**. **Best practices** * Use entities to extract important values from unstructured input. * Provide clear **rules or regex patterns**. * Add **examples** to help the AI recognize the entity. *** ### How to Create an Entity Go to the **Entities** section. Click **Add Entity** to begin. * **Name** — e.g., “UTR Number” * **Rule** — pattern/format of the data (e.g., regex for transaction numbers) * **Description** — short explanation of what the entity represents * **Examples** — add **positive** and **negative** examples in sentences Entities help AI agents automatically capture and use structured values like order IDs or phone numbers during conversations. *** # Managing Human Users Source: https://guides.robylon.ai/human-users/add-human-users The Human Users tab is where you manage the people who can access the Robylon platform. Here, you can view all active users, assign roles, and control permissions. ## Roles & Permissions Each user is assigned a role that defines their level of access: | Role | Chats & Tickets | Settings Access | API Keys | Billing | | --------------- | --------------------------------- | ------------------- | -------- | ------- | | **Agent** | Only their assigned chats/tickets | ❌ | ❌ | ❌ | | **Supervisor** | All chats and tickets | ❌ | ❌ | ❌ | | **Admin** | All chats and tickets | ✅ (settings access) | ❌ | ❌ | | **Super Admin** | Everything | ✅ | ✅ | ✅ | Use the **least privilege principle**—assign only the minimum role needed for each user’s job. *** ## Add a New Human User Go to **Teams → Human Users** in the sidebar. Press the **Add New User** button. Provide the **name** and **email** of the new user. * Select which **channels** (chat, voice, tickets) the user can access. * Assign one of the roles: **Agent**, **Supervisor**, **Admin**, or **Super Admin**. The user will receive an **activation email** at the provided address. *** ## Best Practices * Assign **Agent** roles to frontline staff handling conversations. * Use **Super Admin** roles sparingly (e.g., senior platform owners). * Rotate or disable inactive accounts to maintain security. * Review user roles quarterly to ensure compliance. *** # Human user statuses Source: https://guides.robylon.ai/human-users/agent-status The Agent Status tab allows you to configure the availability states that agents can have on the Robylon platform. These statuses determine whether an agent is eligible to handle chats, tickets, or calls. ## Default Statuses By default, the following statuses are available: * **Online** — Agent is active and available to take queries. * **Offline** — Agent is logged out or unavailable. * **Unavailable** — Agent is logged in but not accepting new queries. *** ## Adding a New Status Open the **Agent Status tab** and select **Add New Status**. Enter a clear name for the new status (e.g., **Break**, **In Training**, **On Call**). Assign a **symbol or icon** to visually identify the status. Select whether an agent in this status is considered **available to handle queries**. Save the status so it appears in the list of agent states. *** ## Best Practices * Keep statuses **simple and descriptive** (e.g., "Lunch Break" vs. just "Unavailable"). * Use **non-available statuses** for internal tracking (e.g., “Training” or “Meeting”). * Standardize across teams to avoid confusion when monitoring queues. Statuses can be used in **analytics and assignment logic** to better understand agent workload and availability. *** # Human agent assignment logic Source: https://guides.robylon.ai/human-users/assignment The Assignment Logic tab controls how chat and voice agent handovers are handled. You can configure rules for agent assignment, working hours, out-of-office handling, and holiday schedules. ## Chat Assignment Settings ### Assignee Group Choose which roles can handle chats: * **Agents** * **Supervisors** * **Admins** * **Super Admins** ### Assignment Logic Define how chats are assigned to human agents: * **Manual** — Supervisor/admin assigns chats directly. * **Round Robin** — Chats rotate evenly among available agents. * **Balanced** — Chats go to the agent with the **least pending tickets**. You can also set the **maximum number of tickets** that can be auto-assigned to a single agent. ### Auto Salutation Automatically send a **welcome message** when a chat is assigned to a human agent. ### Working Hours Configure when human agents are available: * Set **working days** of the week * Define **working hours** per day ### Out of Office Settings Decide how chats are handled outside working hours: * Assign chats to humans during **out-of-office hours** (yes/no) * Allow **verification tasks** to be assigned after hours (yes/no) * Define the **message shown to users** when no agents are available ### Holidays Add public holidays where agents will not be available. Chats received during these days follow the **out-of-office settings**. *** ## Voice Call Assignment Settings ### Call Configurations * Create **voice call configurations** with a **name** and a **list of phone numbers**. * These numbers will receive calls when a **voice handover** occurs. ### Working Hours Define the working days and hours for **voice agents**, similar to chat. ### Out of Office Settings Set how calls are handled when agents are unavailable: * Decide whether calls should be routed to humans outside working hours. * Add a **message** to be played when no agents are available. ### Transfer Types * **Warm Transfer** — Call is transferred with a **handover message**. * **Cold Transfer** — Call is transferred **without any message**. Use **Warm Transfer** for customer-facing teams to ensure smoother context-sharing. Use **Cold Transfer** for urgent or high-volume scenarios. *** # Tags & Canned Responses Source: https://guides.robylon.ai/human-users/tags-canned The Agent Productivity tools include Tags and Canned Responses, which help agents classify conversations and respond more efficiently. ## Tags ### What are Tags? **Tags** are custom identifiers that human agents can add to chats for classification and tracking. They help in organizing conversations, reporting, and analytics. ### How to Use Tags * Apply tags while handling a conversation to classify it (e.g., *Billing Issue*, *Refund*, *Technical Support*). * Tags can be viewed and managed under the **Tags section** in the **Agent Productivity tab**. * Multiple tags can be applied to a single conversation for precise categorization. Use a **consistent taxonomy of tags** across teams for accurate reporting and analytics. *** ## Canned Responses ### What are Canned Responses? **Canned Responses** are preset replies that agents can quickly insert into chats, reducing response time and ensuring consistency. ### How to Use Canned Responses * In a chat window, type **`/`** followed by the command to insert a canned response. * Example: typing `/refund_policy` inserts the full refund policy text. * This allows agents to provide **faster, standardized responses** to common queries. ### Adding a New Canned Response 1. Go to the **Canned Responses section** under Agent Productivity. 2. Click **Add a Canned Response**. 3. Define the **command** (shortcut text) and the **response** (full reply). 4. Save — agents can now use it immediately in chats. Canned responses should be reviewed and updated regularly to reflect the latest policies and messaging tone. *** # Robylon Guides: Introduction Source: https://guides.robylon.ai/index Get an overview of Robylon's AI agents - how they work, and what you can automate. Learn key concepts & terminologies. ### **What is Robylon?** Robylon is a no-code platform designed to build and deploy autonomous AI agents for your business. Unlike simple chatbots that only answer questions, Robylon agents can perform tasks, follow complex instructions, and integrate with your existing tools to automate entire business processes. ## What are AI Agents? AI agents are autonomous software tools that perform tasks, make decisions, and interact with their environment intelligently and rationally. They use artificial intelligence to learn, adapt, and take action based on real-time feedback and changing conditions. AI agents can work on their own or as part of a bigger system, learning and changing based on the data they process. ## How Does Robylon Work? Building a powerful AI agent with Robylon is a straightforward process that doesn't require any coding expertise. The core workflow can be broken down into four simple steps: 1. **Build the Brain (Knowledge Base):** You start by creating a centralized knowledge base for your agent. This is where you upload all the information you want to know. You can connect various data sources like documents (PDFs, DOCX), spreadsheets, entire websites, and more. This knowledge forms the foundation of your agent's intelligence. 2. **Define the Actions (Workflows & Tools):** This is where Robylon's true power lies. You design **workflows**—step-by-step instructions that tell your agent what to do. You can also connect it to **tools** like your CRM, calendar, or internal databases via APIs, giving it the ability to perform real-world actions. 3. **Create Your AI Agent:** You assemble your agent by combining the knowledge base (what it knows) with the specific workflows and tools it should have access to (what it can do). 4. **Deploy & Interact:** Once your agent is ready, you can deploy it across multiple channels. Embed it as a widget on your website, integrate it into Slack or WhatsApp, or use it via an API to power your internal applications. ## Why Robylon? | | **Normal AI Agent** | **Robylon AI Agent** | | ----------------------- | ----------------------------------------------------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Primary Goal** | To answer questions. | To understand intent, reason, and **execute tasks**. | | **Core Components** | A knowledge source. | A knowledge base **+ Workflows and Tools.** | | **Example Interaction** | User: "What are your business hours?" Bot: "9 AM to 5 PM." | User: "I need to book a demo for next week." Robylon Agent: "Sure, I see two open slots. Which one works for you? I will then create the event and add it to our CRM." | Essentially, an AI Agent is an autonomous entity that can perform tasks on behalf of your business, interacting with users and systems to achieve a specific goal. ## Key Features of Robylon * **Autonomous Workflows:** Design multi-step, conditional logic that guides your agent through complex processes without human intervention. * **Centralized Knowledge Base:** Connect multiple data sources (files, websites, text) into a single, cohesive brain for your agents. * **Tool & API Integration:** Give your agents real-world capabilities by connecting them to thousands of third-party applications and internal databases. * **Advanced AI Models:** Leverage the power of leading models like GPT-4o, Gemini 2.5 Pro, etc., to ensure your agents are intelligent, conversational, and accurate. * **Multi-Channel Deployment:** Deploy your agents where your users are—on your website, Slack, WhatsApp, and more. **Analytics & Insights:** Monitor your agent's performance, track conversations, and gather valuable insights to continuously improve their effectiveness. # Robylon × Bubble Integration: No-Code Setup Guide Source: https://guides.robylon.ai/integrations/bubble Connect Robylon to your Bubble app in minutes. Add chat widgets, secure webhooks, and event tracking-without code. Bubble is a no-code website builder. ### Step 1: Generate and Copy Your Robylon Chatbot Embed Code. 1. Log in to your Robylon account and open the dashboard. 2. In the left panel, click on the settings button to navigate to the settings page. Robylon dashboard with the left sidebar open and Settings selected to configure your chatbot 3. Within the settings tab, select [Channels](https://app.robylon.ai/settings/channels). Settings → Channels screen in the Robylon dashboard for bot integrations 4. Click on the [Chatbot](https://app.robylon.ai/settings/channels/chatbot/interface) widget to access the chatbot details. Channels page with the Chatbot widget selected to access bot details 5. Navigate to the [Install](https://app.robylon.ai/settings/channels/chatbot/install) tab within the chatbot details page. Chatbot details showing the Install tab for adding the widget to Bubble 6. Select JavaScript to embed code in Bubble. Install tab with the JavaScript embed option chosen for Bubble apps 7. Copy the provided chatbot embed code. Robylon chatbot Bubble integration embed code snippet in the dashboard ### **Step 2**: Embed Robylon Chatbot on Your Bubble App 1. Once you’ve copied your Robylon embed code, sign in to your Bubble account and head to your account dashboard. 2. On your dashboard, pick the Bubble app or website you wish to embed the chatbot on. Bubble dashboard to pick the app or page where the Robylon chatbot will be embedded 3. Scroll down to the section of the page you want to add the embed code. 4. On the left sidebar of the editor, locate the HTML component and drag it to any section of the page to add a chatbot. Bubble editor: drag the HTML element from the left sidebar onto the page for Robylon chatbot integration 5. Double-click on the HTML component to reveal the code editor. Double-click the HTML element in Bubble to open the code editor 6. Paste the Robylon Embed code copied earlier into the editor. Paste the Robylon embed code into the Bubble HTML editor 7. You can now preview your Bubble app to test your chatbot. Preview the Bubble app to test the Robylon chatbot integration Page preview showing the Robylon chatbot launcher visible on the site Chat window open, confirming the Robylon chatbot works inside the Bubble app # Robylon × Framer Integration: Add Chat to Your Site Source: https://guides.robylon.ai/integrations/framer Embed Robylon on Framer sites with copy-paste code. Configure theme, launcher behavior, and identity for personalized conversations. ### **Step 1: Generate and Copy Your Robylon Chatbot Embed Code.** 1. Log in to your Robylon account and open the dashboard. 2. In the left panel, click on the settings button to navigate to the settings page. Robylon dashboard settings page showing where to access embed settings for chat widget 3. Within the settings tab, select [Channels](https://app.robylon.ai/settings/channels). Robylon dashboard, highlighting Channels section to access chatbot integration options 4. Click on the [Chatbot](https://app.robylon.ai/settings/channels/chatbot/interface) widget to access the chatbot details. Chatbot widget settings in the dashboard for integration with Framer 5. Navigate to the [Install](https://app.robylon.ai/settings/channels/chatbot/install) tab within the chatbot details page. Click on the Install tab in the Robylon chatbot panel, select the embed code method for Framer 6. Select React to embed code in Framer. Embed configuration screen showing React option for Robylon chatbot in Framer 7. Copy the provided chatbot embed code. Robylon React embed code snippet ready for copying to the Framer site ### **Step 2**: Embed the code in Framer to Display Your Chatbot Widget  1. To embed the code in Framer, log in to your Framer account and go to your dashboard. 2. Click on the website and select the page on that website you want to add the Robylon chatbot to. Framer dashboard with project selection view 3. Now, click on the settings icon in the top right part of your Framer website dashboard. The Framer site settings menu, where custom code can be added 4. Click on the ‘General’ option in the sidebar of the Settings page. Scroll down to the ‘Custom Code’ section in settings. 5. Now paste the copied code of Robylon bot here at the End of the head tag. Paste the Robylon embed code into the Framer custom code head section 6. Then save the settings and publish your Framer website to see the chat icon on that page. Framer settings with Custom Code fields and a highlighted Publish button to activate the Robylon chatbot 7. Test the chatbot to see if it works fine, then you can push it for production. # Robylon × Webflow Integration: Add Chat to your site Source: https://guides.robylon.ai/integrations/webflow Install the Robylon chat widget on Webflow. Add chat widgets, secure webhooks, and event tracking-without code. ### **Step 1: Generate and Copy Your Robylon Chatbot Embed Code.** 1. Log in to your Robylon account and open the dashboard. 2. In the left panel, click on the settings button to navigate to the settings page. Robylon dashboard showing the Settings menu to start Webflow chat widget setup 3. Within the settings tab, select [Channels](https://app.robylon.ai/settings/channels). Robylon Settings → Channels screen highlighting Chatbot channel selection 4. Click on the [Chatbot](https://app.robylon.ai/settings/channels/chatbot/interface) widget to access the chatbot details. Robylon Chatbot widget page where the Install tab is accessed 5. Navigate to the [Install](https://app.robylon.ai/settings/channels/chatbot/install) tab within the chatbot details page. Robylon Install tab set to JavaScript embed for Webflow integration 6. Select JavaScript to embed code in Webflow. Robylon5js PnJavaScript embed code option selected in Robylon for Webflow embed 7. Copy the provided chatbot embed code. Copy the embed code snippet for Robylon chatbot to use in Webflow pages ### **Step 2**: Embed the code in Webflow to Display Your Chatbot Widget  1. To embed the code in Webflow, log in to your Webflow account and go to your dashboard. 2. Hover over the website you want to add the chatbot to and click on “Open in Webflow”. On the Webflow dashboard listing’s workspace, select which site to embed the Robylon chatbot 3. On the designer page, click on the file icon (Pages) in the left navbar of the Webflow site designer. Webflow page settings icon, opening the custom code area for chatbot embed 4. Hover over the page you want to embed the chatbot on and click on the settings icon of that page. In the pages panel, select the Edit Page Settings gear icon to configure the Home page before adding Robylon chatbot script 5. Scroll down in the settings panel of that page till you reach the Custom code section. Paste the Robylon chatbot copied code in the custom code section inside the head tag. Then save the changes. Webflow page’s Custom Code section with "Paste embed code in head tag" instruction 6. Publish your website for staging to see the chat icon on that page. Webflow publish settings screen showing staging and production domains for deploying Robylon chatbot integration 7. Test the chatbot to see if it works fine, then you can publish it for production. Webflow site with Robylon chatbot launcher icon visible in the bottom-right corner after integration Webflow post-publish preview showing Robylon chatbot launcher on the site # Robylon × Wix Integration: Add Chat to your site Source: https://guides.robylon.ai/integrations/wix Install the Robylon chat widget on Wix. Add chat widgets, secure webhooks, and event tracking-without code. ### **Step 1: Generate and Copy Your Robylon Chatbot Embed Code.** 1. Log in to your Robylon account and open the dashboard. 2. In the left panel, click on the settings button to navigate to the settings page. Robylon dashboard showing the settings menu for chatbot integration in Wix 3. Within the settings tab, select [Channels](https://app.robylon.ai/settings/channels). Robylon Channels screen with Chatbot channel ready for integration 4. Click on the [Chatbot](https://app.robylon.ai/settings/channels/chatbot/interface) widget to access the chatbot details. Robylon dashboard with Chatbot option highlighted under “Recommended” channels for proactive support and guided help 5. Navigate to the [Install](https://app.robylon.ai/settings/channels/chatbot/install) tab within the chatbot details page. Robylon Chatbot details and the Install tab for embed configuration 6. Select React to embed code in Wix. Robylon dashboard Chatbot Details, Install tab with React option selected, and live chatbot preview of user interaction 7. Copy the provided chatbot embed code. Copy the embed code snippet for Robylon chatbot to paste into the Wix site ### **Step 2**: Sign in to Your Wix Account and Embed Your Chatbot 1. Login to your Wix account and go to the dashboard. 2. On your dashboard, locate and click on the ‘Design Site’ button in the top right corner of the dashboard. Wix Editor site dashboard selecting page to insert Robylon chatbot 3. Select the site you want to add the Robylon chatbot to. Click and load that site into your Wix editor. 4. Scroll down to any section of the page you want to add the Robylon chatbot. 5. Click on the Big Plus icon (Add Elements) on the left sidebar of your Wix editor. Wix Editor with ‘Add → Embed Code’ (HTML iframe) option in site menu 6. Scroll down to locate and click on the ‘Embed Code’ button, then click on the ‘Marketing Tools’ button and select ‘Custom Code’ in that section. Wix HTML iframe settings with Robylon embed code pasted 7. The custom code widget should pop up, click ‘+ Add Custom Code’ in the top right corner of that widget. Wix site publish dialog showing the domain and publish button after integration 8. Paste the copied code snippet into the custom code editor and provide a name for your code. 9. Choose an option under ‘Add Code to Pages’. 10. Click where you want to place your chatbot under the  ‘Place Code’ section. 11. Click Apply and preview your website, and there you should see the floating Robylon chat icon on your website. # Robylon × Wordpress Integration: Add Chat to your site Source: https://guides.robylon.ai/integrations/wordpress Install the Robylon chat widget on Wordpress. Add chat widgets, secure webhooks, and event tracking-without code. ### **Step 1: Generate and Copy Your Robylon Chatbot Embed Code.** 1. Log in to your Robylon account and open the dashboard. 2. In the left panel, click on the settings button to navigate to the settings page. The Robylon dashboard Settings page, where you start WordPress chatbot embed setup 3. Within the settings tab, select [Channels](https://app.robylon.ai/settings/channels). Robylon Channels menu in settings to select Chatbot for WordPress integration 4. Click on the [Chatbot](https://app.robylon.ai/settings/channels/chatbot/interface) widget to access the chatbot details. Chatbot widget tile selected in Robylon Channels to view its settings 5. Navigate to the [Install](https://app.robylon.ai/settings/channels/chatbot/install) tab within the chatbot details page. Robylon Chatbot “Install” tab showing embedding options for WordPress 6. Select JavaScript to embed code in WordPress. JavaScript embed option selected in Robylon for WordPress chatbot code 7. Copy the provided chatbot embed code. Embed code snippet in Robylon ready to copy for WordPress site ### **Step 2**: Install the WPCode plugin on your WordPress Website to embed Robylon’s bot code 1. Login to your WordPress dashboard, and go to the plugin button in the admin bar on the left side, and then click on the ‘Add New Plugin’. WordPress dashboard showing Plugins → Add New screen to install WPCode 2. Then search for the ‘WPCode’ plugin and install it to embed custom code. WPCode plugin listing in the WordPress Add Plugins page is ready for installation 3. Click on the ‘Activate’ button in the plugin. Upon activation, the plugin will be added to the ‘Installed Plugins’ page. Activate the WPCode plugin in the WordPress Installed Plugins list 4. Click on the ‘Installed Plugins’ button in the WordPress admin bar to go to the plugin page. ### **Step 3**: Embed the code in WordPress to Display Your Chatbot Widget  1. Click on the ‘Code Snippets’ labeled menu item in your WordPress admin bar, and it will open the ‘All Snippets’ page. The WordPress sidebar “Code Snippets” menu is visible to embed Robylon code 2. Then, click on the ‘Add New’ button, and this will bring you to the add snippet page. Add a New Snippet page in WordPress showing options to create a custom snippet 3. Click on the ‘+Add Custom Snippet’ button. WordPress WPCode plugin screen with the option to add a custom code snippet for embedding the Robylon agent 4. Then, choose the code type from the list of options that appear on the screen, and click on JavaScript in our case. WordPress snippet creation with a prompt to choose the JavaScript snippet type 5. Now, enter a title for the custom code snippet and paste the copied code for the chatbot, and save the snippet. WordPress snippet editor where you paste the Robylon chatbot code 6. Then you will see 2 options to insert, you can select the ‘Auto Insert’ method, and the snippet will be automatically inserted and executed on your site. WordPress snippet settings showing “Auto Insert” method selected 7. Now you can preview your site to see the embedded Robylon’s chatbot there. WordPress website with Robylon chatbot launcher icon visible in bottom right corner after integration. WordPress site with Robylon Support Bot open, showing chatbot greeting message after successful integration # Click to Improve Source: https://guides.robylon.ai/knowledgebase/click-to-improve The **Click to Improve** feature allows you to audit and refine your AI's performance directly from past conversations. This feature helps you identify which source the AI used to generate an answer and allows you to instantly create a new, corrected Q\&A pair to improve future accuracy. # Overview When reviewing chat history, you may find instances where the AI gave a generic answer or slightly misinterpreted a query. Instead of manually going to the Knowledge Base to create a new rule, you can do it directly from the transcript. With this feature, you can: * **Verify Sources:** See exactly which document or text chunk the AI used to generate the response. * **Edit Content:** Modify the user's question and the AI's response to be more accurate. * **Add to KB:** Save the corrected version as a separate Q\&A entry in your Knowledge Base. # Step-by-Step Guide ### **1. Navigate to Transcripts** From your Robylon dashboard, click on the **Transcripts** tab in the sidebar to view your conversation history. ### **2. Locate the AI Response** Open a conversation and scroll to the specific AI response you wish to audit. You will see a small clickable text labeled **Click to improve** located immediately below the AI's generated message. ### **3. View the Source** Once you click the link, a window will appear displaying the **Source Context**. This highlights the specific section of your uploaded data (PDF, URL, or Text) that the AI relied upon to formulate that answer. ### **4. Edit the Question and Answer** In the same window, you will see fields for **Question** and **Answer**: * **Question:** This is pre-filled with the user's query. You can edit this to make it a standard question. * **Answer:** This is pre-filled with the AI's generated response. You can edit this text to provide the exact, perfect answer you want the bot to give next time. ### **5. Save as a New Q\&A** Once you have refined the text, click the **Add QnA** button in the bottom right corner. This action does not alter your original source document. Instead, it creates a separate Q\&A pair in your Knowledge Base. Moving forward, the AI will prioritize this specific Q\&A pair when asked similar questions, ensuring the improved response is delivered. # Best Practices 1. **Generalize the Question:** When editing the "Question" field, try to remove specific user details (like names or account numbers) so the Q\&A applies to all future users. 2. **Check the Source:** If the source context shown is completely irrelevant, it indicates you may need to upload better documentation regarding that topic to your Knowledge Base. # Custom KB Responses Source: https://guides.robylon.ai/knowledgebase/custom-kb-responses The agents utilize a framework called **Retrieval-Augmented Generation (RAG)**. Instead of relying on the AI's general, pre-existing knowledge (which can lead to hallucinations), RAG forces the AI to "read" the specific knowledge base before answering. # 1. The Foundation: How RAG is Created from the KB Content Here is how the KB content is transformed into a RAG-ready format: * **Parsing:** When you upload content (PDFs, URLs, Help Center articles), the raw text is extracted. * **Chunking:** The AI cannot read an entire manual at once efficiently. We break the documents down into smaller, logical, overlapping paragraphs called "chunks." * **Embedding:** We use advanced machine learning models to translate these text chunks into complex numerical arrays called vectors. * **Storage:** These vectors are stored in a specialized Vector Database. Think of this database as a massive, highly organized library where information is grouped by its underlying meaning, not just its keywords. # 2. Query Expansion: Understanding the Customer's True Intent Customers rarely ask perfect questions. A user might type "reset" or "it won't turn on." If we searched the KB for exactly those words, we might get poor results. To fix this, the agent performs **Query Expansion**. **The Meaning:** Query expansion is the process of taking the user's raw input and enriching it with context from the ongoing conversation to create a highly detailed search query. **The Process:** Before searching the database, our AI analyzes the user's message alongside the chat history. # 3. Semantic Search on Chunks Once the query is expanded, the agent needs to find the exact chunks of data in the Knowledge Base that contain the answer. We do this using **Semantic Search**. **The Meaning:** Unlike traditional search engines that look for exact keyword matches (e.g., matching the word "billing" to "billing"), semantic search looks for meaning. It knows that "pricing," "cost," and "invoice" are related to "billing." **The Process:** The expanded query is converted into a numerical vector (just like the KB chunks were). The system then scans the Vector Database to find the chunks that are mathematically closest in "meaning" to the query. It retrieves a preliminary list of the most relevant chunks. # 4. Reranking of Chunks: Quality Control Semantic search is incredibly fast, but it can sometimes retrieve chunks that are only slightly related to the topic. To ensure the AI gets only the highest-quality information, we apply a **Reranking** process. **The Process:** A secondary AI model acts as a judge. It takes the user's query and strictly evaluates the list of chunks retrieved during the semantic search. It scores each chunk based on how directly it answers the specific question, discarding the "fluff" and sorting the remaining chunks from most relevant to least relevant. # 5. Final KB Response Generation (and Custom Instructions) This is where the magic happens. Now the agent has the perfectly reranked chunks of information, it's time to talk to the customer. The system creates a comprehensive "prompt" package to send to the final Large Language Model (LLM). This package includes: 1. The User's Question. 2. The Top Reranked KB Chunks (the facts). 3. The Custom KB Response Instructions. ### **The Role of Custom Instructions** The Custom KB Instructions act as the "director" of the AI. Even with perfect facts, the AI needs to know how to speak. The instructions dictate the agent's tone, formatting, and constraints. For example, the custom instructions might state: * "Always reply in a friendly, casual tone." * "If the answer involves multiple steps, always use bullet points." * "If the answer is not in the provided text, apologize and offer to connect them to a human agent." The LLM reads the facts, applies the Custom Instructions, and generates the final, perfectly tailored response. # 6. Fine-Tuning the Experience: Chunk Limits & Follow-Up Questions As you configure the agent, you will encounter a few advanced settings that give you ultimate control over the user experience: ### **What is a 'Chunk Limit'?** The chunk limit is the maximum number of text chunks the AI is allowed to "read" when generating a response. **Why it matters:** Feeding the AI too many chunks can cause "information overload," leading to confused answers or higher latency (slower response times). Setting a proper chunk limit (e.g., 3 to 5 chunks) ensures the AI stays laser-focused on the most relevant facts. ### **Follow-Up Questions** A great support agent anticipates what the customer needs next. Based on the chunks retrieved and the response generated, KB agent can automatically generate contextual Follow-Up Questions. # Knowledge Base Source: https://guides.robylon.ai/knowledgebase/index Getting Started with Your Robylon AI Knowledge Base! Welcome to your Robylon AI Knowledge Base! Think of this as the brain of your AI agent, powering it to provide instant, accurate, and consistent answers to all user queries. Everything you add here - documents, Q\&As, and web pages - will be used by your agent to learn and provide accurate, instant answers to user queries. A well-structured knowledge base is the key to a powerful and effective AI assistant. This guide will walk you through everything you need to know about building and managing your knowledge base. ## Adding Information to Your Knowledge Base Your knowledge base is where you store all the information you want your AI agent to know. You can add content in several ways. To get started, navigate to the Knowledge section from the left-hand menu, click on Source and then click on the Add Source button in the top right. You'll see four options to add content: This is perfect for adding specific information, FAQs, or correcting your AI on certain topics. 1. Select 'Add a Q\&A'. 2. Title: Give a clear title to your Q\&A entry for easy reference. 3. Question: Enter a question that you anticipate your users will ask. You can add multiple variations of the same question by clicking + Add question (e.g., 'How do I reset my password?', 'Password reset steps?'). 4. Answer: Provide a clear and concise answer. 5. (Optional) Reference Link: If the answer is sourced from a specific webpage, add the URL here. This helps the AI provide a source for its answers or provides users with a link for more information. 6. (Optional) KB Tags: Add tags to categorize your Q\&A. This can be useful for organizing and filtering your knowledge base. 7. Click Add Q\&A to save. You can upload existing documents to quickly build your knowledge base. This is ideal for policy documents, product manuals, or internal guides. 1. Select 'Upload a File'. 2. You can either drag and drop your file into the window or click Select from Device to browse and choose a file. Supported file types, such as .pdf, .txt, and .docx, are listed within the upload window, with a size limit of up to 50MB. 3. (Optional) KB Tags: Add tags to organize the uploaded file. 4. Click Add File(s). The file will be processed and added to your knowledge base. Sync from Google Drive to add your files. The initial sync will pull the current version of the file. To see new changes later, do a manual refresh or turn on auto-refresh. 1. Select 'Add from Google Drive'. 2. Click Connect to link your Google Drive account. You'll need to grant Robylon AI permission to access your files. 3. Once connected, Select a File by searching or browsing your Drive. You can also choose to Upload the entire folder – ideal for syncing entire documentation libraries at once. 4. Refresh: Set a frequency (e.g., Monthly, Weekly, Daily) for Robylon AI to automatically check for updates to this file. Note that this functionality is available only on Business and Enterprise plans. 5. (Optional) KB Tags & Reference Link: Add tags for organization or a link for reference. 6. Click Add File(s) to upload the content. Pull content directly from your website - pages, FAQs, help center, etc. Keep it current with a manual refresh or a scheduled auto-refresh. 1. Select 'Sync webpage(s)'. 2. Website URL to Sync: Enter the URL of the website you want to crawl. Robylon AI will search for all pages on this site. 3. URL paths: You can control which pages are included or excluded. These allow you to precisely control the scope of the sync, ensuring only relevant content is ingested and avoiding unnecessary pages like login forms. 1. URL paths to include: Use this to sync only specific sections of a website (e.g., /help or /blog). 2. URL paths to exclude: Use this to prevent specific sections from being synced (e.g., /login or /careers). 4. Advanced Settings (Optional): 1. Additional URLs: Add comma-separated URLs of specific pages you want to sync. 2. Wait to load CSS selector: Use a CSS selector to delay scraping for pages with dynamic content (e.g., #content-loaded). Consult your web developer or use browser developer tools to find appropriate selectors. 3. Use XML Sitemap: Enable this to use the website's sitemap for more efficient crawling. 5. Click Crawl. Once the crawling process is complete, a list of discovered URLs will appear. From this list, you can then select which URLs to include or exclude from your knowledge base. 6. Click Publish to add the selected URLs’ content to your knowledge base. Set rules, tag content, and guide the KB Agent. # Knowledge Base from Humans Source: https://guides.robylon.ai/knowledgebase/kb-from-humans The **KB from Humans** feature allows you to automatically generate Q\&A pairs for your Knowledge Base directly from closed human agent conversations. Instead of manually creating Q\&As from scratch, this feature learns from your best agents' responses and converts them into reusable knowledge that your AI chatbot can use to answer similar questions in the future. When enabled, the system monitors closed conversations of selected human agents and generates Q\&A entries from those interactions. You can choose to have these Q\&As go live immediately or require manual review before they are added to the chatbot's active knowledge. # Key Concepts **Auto-Generation from Closed Chats:** Q\&As are created automatically when a monitored human agent's chat is closed. There is no manual trigger required — the system processes conversations as they close. **Agent Selection:** You choose which human agents' conversations should be used for Q\&A generation. Only closed chats from the selected agents are processed. **Auto-Review vs. Manual Review:** You can enable auto-review, which directly adds generated Q\&As in a live (Enabled) state. If auto-review is turned off, generated Q\&As will be in a "Pending for Review" state and must be manually approved before the chatbot uses them. **Maximum Q\&A Limit:** You can set a cap on the total number of Q\&As to be generated, up to a maximum of 500. **Duplicate and Conflict Prevention:** The system checks for duplicates and contradictions with existing KB entries. If a conflict or duplicate is detected, the new Q\&A is discarded and the existing entry is preserved. **Priority:** Generated Q\&As maintain the same high priority as other Q\&As in the Knowledge Base. # Enabling Q\&As from Human Agents ### **Step-by-Step Guide** 1. Log in to your Robylon dashboard. 2. From the left sidebar, navigate to **Knowledge Base**. 3. Click the **Add Source** button. The **Select Source Type** popup will appear with all available source options. 4. Select **Q\&As from Human Agent** at the bottom of the list. If this source has not been enabled yet, you will see a "Not Enabled Yet" badge next to the option. 5. A configuration popup will open, titled **Q\&As from Human Agent**, with the following settings: ### **Configuration Fields** **Conversations from selected human agents:** Use the multi-select dropdown to choose which human agents' closed conversations should be used for Q\&A generation. You can select one or more agents. Only closed chats from these agents will be processed. **Enable auto-review of Q\&As?:** Toggle this setting on or off. When enabled, generated Q\&As will be directly added to the KB in a live state for the chatbot to use immediately. When disabled, Q\&As will be created in a "Pending for Review" state and will require manual approval. **Max Q\&As to be generated:** Enter the maximum number of Q\&As you want the system to generate. This is an integer field with a maximum value of 5,000. The system will stop generating new Q\&As once this limit is reached. # Starting and Stopping Generation When configuring for the first time, you will see two buttons at the bottom of the popup: * **Save** — Saves the configuration without enabling generation. * **Enable** — Saves the configuration and starts Q\&A generation immediately. Once generation has been enabled, a dedicated folder named "Q\&As from Human Agents" will appear in your KB Sources list. A confirmation toast message will appear at the bottom of the screen confirming that the feature has been enabled. # Managing Generated Q\&As ### **The Q\&As from Human Agents Folder** Once enabled, a folder named "Q\&As from Human Agents" will appear as the first entry in your KB Sources page. This folder contains all auto-generated Q\&A entries. The folder view shows a table with the following columns: * **Name** — The name of the generated Q\&A, along with any KB Tags attached to it. * **Human User** — The human agent whose conversation was used to generate the Q\&A. * **Status** — Either "Pending for Review" or "Reviewed." * **Usage Count** — How many times this Q\&A has been used by the chatbot. * **Updated** — The timestamp of the last update. The folder status on the KB Sources page will show as "Processing" if there are any Q\&As still pending review, or "Enabled" if all Q\&As have been reviewed. ### **Toolbar Actions** At the top of the folder view, you have access to the following tools: * **Filters** — A Human Agent filter (multi-select) and a Status filter to narrow down the Q\&A list. * **Search** — A search bar to find Q\&As by name. * **Mark as Reviewed** — Bulk-approve selected Q\&As that are pending review. This button is enabled only when at least one pending Q\&A is selected. * **Download** — Download selected Q\&As. Enabled when at least one Q\&A is selected. * **Delete** — Remove selected Q\&As. Enabled when at least one Q\&A is selected. You can select individual Q\&As using the checkbox on each row, or use the header checkbox to select all. ### **Q\&A Settings** Inside the folder, you can access the Q\&A Settings panel by clicking the **Q\&A Settings** button in the top-right corner. This opens a side panel where you can modify the generation configuration at any time. The settings panel shows the current configuration, including the selected agents, auto-review toggle, and maximum Q\&A limit. An "Enabled" badge is shown in the top-right corner when generation is active. If generation is currently active, you can adjust the settings and click **Save** to apply changes, or **Stop** to halt Q\&A generation. If generation is currently stopped, you will see **Start** and **Cancel** buttons instead. Stopping generation does not delete any existing Q\&As in the folder. It only pauses the creation of new ones. # Reviewing Individual Q\&As Clicking on any row in the folder opens a detail popup for that specific Q\&A. This popup allows you to review and edit the generated content. ### **Editable Fields** * **Name** — A short title for the Q\&A entry. * **Question(s)** — The question or set of questions the Q\&A covers. * **Answer** — The response the chatbot will use when this Q\&A is matched. * **Custom Instructions** — Any specific instructions for how the chatbot should use this Q\&A. * **Details** — Includes KB Tags and a reference link to the source conversation. ### **Review Actions** When a Q\&A is in the "Pending for Review" state, the popup will display: * **Save for Later** — Saves any edits you make without changing the review status. * **Discard** — Permanently deletes the Q\&A. A confirmation popup will appear before deletion. * **Mark as Approved** — Approves the Q\&A, changing its status to "Reviewed" and making it available for the chatbot to use. Once a Q\&A has been reviewed and approved, a **Source** field will be added to its details, showing which human agent's conversation was used to generate it. # Important Notes * You cannot manually add Q\&As inside the "Q\&As from Human Agents" folder. All entries in this folder are auto-generated. * The folder can be deleted from the KB Sources page, which will remove all generated Q\&As. * The download button on the KB Sources page works on this folder and will export all Q\&As within it. * The source type for this folder is "Q\&A," and it will appear when you apply the Q\&A filter on the KB Sources page. * Generated Q\&As are not currently included in the memory and storage system. This is planned for a future phase. # Best Practices **Start with Your Best Agents:** Select agents who consistently provide accurate, helpful, and well-structured responses. The quality of generated Q\&As depends directly on the quality of the source conversations. **Use Manual Review Initially:** When first enabling the feature, keep auto-review turned off. This allows you to inspect the quality of generated Q\&As and build confidence in the system before enabling auto-review. **Set a Conservative Q\&A Limit:** Begin with a lower limit (e.g., 100–500) and increase it once you are satisfied with the quality and relevance of the generated Q\&As. **Review Regularly:** Check the folder periodically and approve or discard pending Q\&As. Letting too many Q\&As pile up without review reduces the value of the feature. **Edit Before Approving:** Use the review popup to refine questions and answers. Remove specific customer details such as names, order IDs, or account numbers so that the Q\&A applies broadly to all future users. # Managing Your Knowledge Base Source: https://guides.robylon.ai/knowledgebase/managingkb Manage sources, tags, usage limits, and activate your KB Agent to deliver accurate, on-brand answers. After adding sources, manage them from the **Knowledge** page. You’ll see: * **Source Name** — file name, Q\&A title, or URL for quick identification * **Status** — *Processing* (ingesting) or *Enabled* (active) * **Source Size** — file size * **Usage Count** — how often the AI used this source * **Updated** — last update time You can click any source to **edit**, **add/modify KB Tags**, or **preview content**. **Pro tip:** Keep titles consistent (e.g., “Policy • Returns • v2”) so they’re easy to find and maintain. *** ## KB Tags **KB Tags** are **key–value pairs** (e.g., `department: billing`, `product: x`, `audience: internal`) that let you **route** and **scope** which sources the AI can use. **Example:**\ For billing queries, workflows can be configured to **only search** documents tagged `department: billing`. Use tags to separate internal vs external docs, products/SKUs, regions, or compliance levels. This keeps answers relevant and reduces noise. * Standardize keys (e.g., `department`, `product`, `audience`) * Keep values short and predictable (e.g., `billing`, `returns`, `pro`) * Use 1–3 tags per source for clarity *** ## Account Usage & Limits Plans include limits on **total sources** and **storage size**. * Monitor current usage at the top of the **Source** page. * If you’re close to limits, plan a cleanup or consider an upgrade. **Housekeeping matters:** Archive or disable stale sources to keep the index lean and fast. *** ## Activate with the KB Agent Open **KB Agent** from the left menu to connect instructions and sources: 1. Add **Custom Instructions** to guide **tone**, **format**, and **decision-making**. 2. Review the list of **active KB Sources** and confirm the scope aligns with your tags and workflows. **Instruction hygiene:** Be explicit about **what to cite**, **what to avoid**, and **how to answer** (bullet points vs paragraphs, add links, etc.). *** ## Q\&As and KB Gap * **QnAs:** Create **canonical** Q\&A pairs for high-traffic questions. * **KB Gap:** See questions the AI **couldn’t answer** with current sources—then fill those gaps with new docs or Q\&As. Review KB Gap weekly, add or tag missing sources, and update instructions if needed. This continuous loop boosts accuracy and deflection. * One clear question per entry * Short, factual answer with links for depth * Consistent voice and formatting *** ## Checklist * Sources added and named consistently * Tags applied (department/product/audience) * Usage & limits reviewed * KB Agent instructions set and tested * Q\&As created for top queries * KB Gap reviewed and actioned weekly **Next step:** Test with 10–15 real customer questions and iterate tags/instructions based on results. # Suggested Messages Source: https://guides.robylon.ai/knowledgebase/suggested-messages Proactive messages allow you to reach out to website visitors before they initiate a conversation. This guide walks you through setting up the message and building the workflow that handles user responses. # How to Set Up Proactive Messages ### **1. Navigate to the Outbound Tab** Log in to your Robylon dashboard and click on the **Outbound** tab in the main navigation menu. ### **2. Create a New Message** Click the **Create a proactive message** button to start configuration. ### **3. Configure Message Content** * **Message Type:** Choose whether you want to display **Text** or a **Button**. * **Content:** Enter the text you want to show to the customer (e.g., "Hi! Need help with pricing?"). ### **4. Set Display Rules** Define when and where the message should appear: * **Timing:** Set the delay time (e.g., send after 5 seconds). * **Targeting:** Specify the **Page URL** where this message should be shown. ### **5. Save** Click **Next** to store your message settings. # Build the Workflow Once the message is set up, you need to configure what happens when a user clicks or replies to it. ### **1. Go to Workflows** Navigate to the **Workflows** section from the **Settings** tab of the dashboard. ### **2. Create a New Workflow** Click **Create Workflow** to start a new automation flow. ### **3. Add a Trigger** Select the trigger step to initiate the flow. Configure it: * **Trigger Category:** Select **Channel Trigger**. * **Channel Type:** Select **Robylon Chat**. * **Trigger Event:** Select **When a user replies to a proactive message**. ### **4. Define the Action** Now, build the journey for the user. You can add actions such as: * Sending a follow-up message. * Asking for customer details (Lead Gen). * Assigning the chat to a human agent. ### **5. Publish Workflow** Once your flow is designed, click **Publish** to make it live. # Bulk Testing Source: https://guides.robylon.ai/knowledgebase/untitled-page The **Bulk** **Testing** feature lets you validate your Knowledge Base at scale. Upload a list of questions as a CSV, let the KB agent generate answers automatically, and then review each response for accuracy — all from a single interface inside the Knowledge Base. # Overview Instead of manually testing your chatbot one question at a time, KB Testing allows you to: * **Bulk-test questions:** Upload dozens or hundreds of questions in one go. * **Auto-generate answers:** The system runs each question through the KB agent and records the AI's response. * **Review and grade:** Inspect each answer, check the sources used, and mark it as positive or negative. * **Export results:** Download the full test as a CSV with questions, answers, grades, and notes. # Accessing KB Testing 1. Log in to your Robylon dashboard. 2. From the left sidebar, navigate to **Knowledge Base**. 3. Click the **Testing** tab below the KB Gap section. The Testing page displays all your tests in a list view with the following columns: * **Name** — The name of the test (clickable to open the individual test view). * **Total Questions** — The number of questions in the test. * **Status** — The current state of the test: **Running**, **Partially Reviewed**, or **Reviewed**. * **Run Date** — The date and time the test was created, along with who ran it. At the top of the page, you will find a **Status** filter dropdown, a **Search** bar to find tests by name, and action buttons for **Download** and **Delete** (enabled when one or more tests are selected). # Creating a Test ### **1. Open the Create Test Modal** Click the **+ Create Test** button in the top-right corner of the Testing page. ### **2. Enter a Test Name** Provide a name for your test in the **Test Name** field. This is a required field. ### **3. Upload Your Questions CSV** Upload a CSV file containing the questions you want to test. You can drag and drop the file or click **Upload CSV** to browse your files. The CSV must contain a **Question** column header. Click the **Download Sample CSV** link to get a template with the correct format. ### **4. Start the Test** Click **Start Test** to begin. The modal closes and your test appears in the list with a **Running** status. The KB agent processes each question in the background and the status updates automatically as answers are generated. # Reviewing a Test Click on any test name in the list to open the individual test view. This page contains the following sections: ### **Stats Widget** A summary bar at the top shows four metrics: * **Total Questions** — The total number of questions in the test. * **Positive Reviews** — The count of answers graded as positive (shown in green). * **Negative Reviews** — The count of answers graded as negative (shown in red). * **Pending Reviews** — The count of answers not yet reviewed (shown in orange). ### **Q\&A List** Below the stats widget, a table lists every question in the test with the following columns: * **Question** — The full question text (clickable to open the review slider). * **Answer Status** — Shows whether the KB agent successfully generated an answer. * **Answer Rating** — The current feedback status: **Pending**, **Good**, **Acceptable**, or **Bad**. You can filter the list using the **Answer Status** and **Rating** dropdowns, or search for specific questions using the search bar. # Evaluating Individual Answers Click on any row in the Q\&A list to open the **Evaluate Answer** slider on the right side of the screen. The slider contains: ### **Question and Answer** The question is displayed at the top, followed by the AI-generated answer. Both fields are read-only. ### **Cited KB Source** An expandable accordion showing the KB sources the AI used to generate the answer. This works the same way as the Click to Improve feature — you can see exactly which documents or text chunks were referenced. ### **Review Answer** Grade the answer using one of the available rating options: **Good**, **Acceptable**, or **Bad**. If you select **Bad**, additional options appear for tagging the reason for failure: **Hallucination**, **Incomplete**, **Wrong source**, or **Other**. At least one reason must be selected. An **Add internal note** text area is always available for adding free-form feedback regardless of the grade. Click **Save** to record your review, or **Cancel** to discard changes and close the slider. # Bulk Actions To speed up the review process, you can select multiple Q\&As using the checkboxes and apply feedback in bulk: * **Mark as Positive** — Applies a positive rating to all selected Q\&As at once. * **Mark as Negative** — Applies a negative rating to all selected Q\&As at once. Bulk actions update the rating and timestamp for all affected Q\&As immediately without opening the review slider. # Downloading Test Results You can download test results in two ways: * **From the Testing page:** Select one or more tests using the checkboxes and click the **Download** icon. Multiple tests are packaged into a zip file. * **From the individual test view:** Click the **Download** icon in the top-right corner of the test page. The exported CSV includes the following columns: **Question**, **Answer**, **Answer grading**, **Reasons for negative grading**, and **Additional notes**. All Q\&As are included regardless of their review status. # Deleting Tests Select one or more tests from the list and click the **Delete** icon. A confirmation popup will appear before the tests are permanently removed. You can also delete an individual test from within the test view using the **Delete** icon in the top-right corner. # Best Practices **Start with a focused question set:** Begin with 50–100 questions covering your most common customer queries. This gives you a manageable first batch to review and helps you quickly identify gaps in your Knowledge Base. **Use the Sample CSV template:** Download the sample CSV to ensure your file has the correct format before uploading. This avoids upload errors and failed tests. **Review negative answers carefully:** When an answer is graded as negative, check the Cited KB Source section to understand why. If the wrong source was used, consider improving your KB content or adding a dedicated Q\&A for that topic. **Export and share results:** Download completed tests and share the CSV with your team to collaboratively identify patterns in incorrect answers and prioritize KB improvements. # Languages Source: https://guides.robylon.ai/languages The Languages section allows you to configure the languages used for customer communication. You can select a default language and choose additional supported languages. The chatbot will be able to understand and respond in the languages selected here. # Steps to Add Languages 1. Log in to your Robylon dashboard. 2. From the left sidebar, click Settings. 3. On the Settings page, under General, select Languages. 4. Under Choose Default Language, select the primary language for your chatbot. 5. Under What languages should your chatbot support?, select the additional languages you want the chatbot to support. 6. Click Save to apply the changes. # Default Language The default language is the primary language used by the chatbot for communication. It is the language in which the chatbot is initially configured to interact with users. If the chatbot cannot detect the user's language or the user interacts in an unsupported language, the chatbot responds using the default language. You can select the default language from the available language options. # Supported Languages The chatbot supports multiple languages for communication. When a user interacts in any of the selected languages, the chatbot responds accordingly. Supported languages include: ### Primary Languages English, Spanish, French, Hindi, Arabic ### Chinese Simplified Chinese, Traditional Chinese ### European Languages Bosnian, Bulgarian, Catalan, Croatian, Czech, Danish, Dutch, Estonian, Finnish, German, German (Formal), Greek, Hungarian, Latvian, Lithuanian, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovenian, Swedish, Ukrainian ### Asian Languages Bengali, Kannada, Japanese, Korean, Malay, Marathi, Mongolian, Tamil, Telugu, Thai, Vietnamese ### Middle Eastern Languages Hebrew, Persian ### Other Supported Languages Brazilian Portuguese, Indonesian, Italian, Swahili, Turkish # Robylon Models Comparison: Pick the Right LLM Source: https://guides.robylon.ai/model-comparison Compare supported models by cost, speed, context length, and quality. See trade-offs and defaults for chat, email, and WhatsApp support. ## AI Model Comparison for Enhanced Customer Support Choosing the right AI model is essential for delivering exceptional customer support experiences with Robylon. Our platform combines advanced AI with continuous human oversight, ensuring your chatbot handles even the most complex queries with accuracy and efficiency. Each model below is evaluated for its technical strength, communication quality, response speed, ability to perform tasks like scheduling or account management, and effectiveness in managing intricate, multi-step interactions. ## Highly Recommended Models ### 1.) Gemini 2.5 Flash Gemini 2.5 Flash is a next-generation AI model engineered for lightning-fast, responsive customer interactions. **Specialized For**: * Real-time, high-volume customer support where speed is critical * Dynamic, multi-step workflows and conversational experiences * Scenarios requiring immediate, accurate responses * Voice-based and interactive support channels ### 2.) Gemini 2.5 Pro Gemini 2.5 Pro delivers premium accuracy and advanced reasoning, making it ideal for mission-critical support scenarios. **Specialized For**: * Complex technical troubleshooting and premium customer support * Regulated industries demand precise, reliable analysis * Multi-step problem-solving with deep contextual understanding * Environments where accuracy and compliance are paramount ### 3.) Claude Sonnet 3.7 Claude Sonnet 3.7, Anthropic’s flagship model, is renowned for its transparent reasoning and sophisticated analytical capabilities. **Specialized For**: * Enterprise-level support requiring detailed, empathetic, and clear explanations * Complex, multi-step technical problem-solving * Scenarios demanding high-quality, context-rich interactions * Applications needing deep domain expertise ### 4.) Claude Sonnet 4 Claude Sonnet 4 represents the latest in Anthropic’s AI innovation, offering even greater reasoning depth and communication clarity. **Specialized For**: * Premium support scenarios with advanced analytical requirements * Multi-step troubleshooting in highly specialized domains * Customer interactions where transparency and detailed reasoning are essential * High-stakes environments need state-of-the-art AI performance ## Other Model Options | **Model** | **Specialized for** | | :---------- | :-------------------------------------------------------------------------------------------------------------------------------------------- | | GPT-4o | - Multilingual, multimodal support  - Fast, empathetic communication - Complex problem solving excels in high-volume, enterprise environments | | GPT-4.5 | - Knowledge-intensive, specialized scenarios - Deep context analysis, resource-intensive and ideal for detailed, specific interactions | | DeepSeek-V3 | - Analyzing extensive data, logs, or complex documentation - Ideal for in-depth, detailed troubleshooting | | DeepSeek-R1 | - Documentation-driven support  - Conversational search and documentation-based interactions | | O3 Mini | - Rapid technical support - Logical troubleshooting benefits from prompt engineering for enhanced empathy | # Dev Components Source: https://guides.robylon.ai/product/dev-notes Dev components bring development and engineering flexibility into workflows. They allow you to call APIs, run Python functions, connect integrations, and annotate workflows with notes. ### API The **API block** sets up external API calls inside workflows. * Pulls or pushes data to a client’s database or third-party service. * Can be invoked directly or by an **AI agent** when additional context is required. * Example uses: * Check a **user’s subscription status** * Retrieve **order details** from an ERP * Update a **ticket property** in a helpdesk system *** ### Function The **Function block** allows embedding custom Python code. * Accepts **specific inputs** and returns **structured outputs**. * Highly versatile for validation, transformation, or computation. * Example uses: * Validate **transaction hashes** * Check if a **URL is valid** * Normalize data (e.g., phone numbers, dates) Keep functions **small and modular**. Promote frequently used logic into a **Tool** for reuse across workflows. *** ### Integration The **Integration block** connects to third-party apps and fetches structured data. * Common integrations: **Google Sheets**, **CRMs**, **Payment processors** * Example uses: * Read/write rows in a Google Sheet * Pull customer info from Salesforce * Fetch product catalog data from Shopify *** ### Notes The **Notes block** adds inline documentation to a workflow. * Helps explain **why a step exists** or what a **condition means**. * Improves collaboration by making workflows easier to audit and maintain. * Notes are **non-executable** and purely for context. *** # Gen AI Components Source: https://guides.robylon.ai/product/gen-ai Gen AI components allow you to build intelligent automation blocks powered by AI. They use prompts, knowledge bases, and tools to generate, evaluate, or extract information from conversations. ### AI Agent The **AI Agent block** creates a dedicated AI agent within a workflow. * **One agent per workflow** is allowed. * The agent is the **core driver** of automation, responsible for handling logic and conversations. * Agents can: * **Call tools** (e.g., APIs, functions) * **Generate replies** for customers * **Fetch information** from the knowledge base Define clear **instructions and goals** for each AI agent to avoid overlap or misrouting across workflows. *** ### Generate The **Generate block** uses AI to create content or validate responses.\ Common uses: * Generate a **reply** based on inputs and KB context * **Evaluate** if a user’s query has been fully answered * Summarize long responses into concise answers * Provide **alternatives** when one reply is insufficient *** ### Extract The **Extract block** uses AI to pull structured details out of user queries, similar to **Capture**, but more flexible.\ Examples: * Extracting an **order ID** from a free-form message * Pulling a **date** or **location** from a user’s text * Identifying **entities** like product names or account numbers Use **Extract** when inputs are **unstructured or varied**. Use **Capture** when the format is **predictable** (e.g., fixed ID patterns). *** # Listen Components Source: https://guides.robylon.ai/product/listen Listen components capture inputs from customers, ticketing systems, APIs, or other workflows. They ensure automations receive the right context before taking action. ### Listen The **Listen block** captures **incoming data**: * User inputs in chat or voice * Data from ticket handling systems * Responses from APIs Use it as the **starting point** for gathering context before moving to logic or actions. *** ### Buttons **Buttons** provide users with a **list of predefined options** to select.\ Examples: * “Check Order Status” / “Request Refund” / “Talk to Agent” * Yes / No confirmations * Quick navigation in chatbots This reduces ambiguity and speeds up intent detection. *** ### Capture The **Capture block** extracts **structured variables** from user responses. * Examples: `order_id`, `transaction_id`, `email` * Can also detect **intent** based on the reply * Useful for **sorting queries** and routing to the right workflow or sub-agent *** ### Upload Image Allows a chatbot to **receive images** from users.\ Common use cases: * Upload **payment confirmation screenshots** * Submit **ID proof** for KYC * Share photos of **defective products** for return requests *** ### Carousel A **Carousel** displays a **scrollable set of cards**—each with an **image, title, description, and button**. * Cards can be created **manually** or from a **JSON list** * Each card button sends its data to the AI agent * Ideal for chatbots in **e-commerce**, **product discovery**, or **FAQ navigation** Carousels work best for **visual menus** (e.g., product recommendations, feature comparisons) where users can click through instead of typing queries. *** # Logic Components Source: https://guides.robylon.ai/product/logic Logic components bring reasoning and control flow to workflows. They let you branch paths, update variables, and introduce delays to handle real-world timing and decision-making. ### Conditions The **Conditions block** works like **if–else statements** in programming. * Diverts workflow paths based on rules (e.g., `if status = open → continue`, `else → close`). * Common use cases: * Route a ticket based on **priority level** * Send premium users to a **VIP workflow** * Trigger escalation if **CSAT \< 3** *** ### Set Variables The **Set Variables block** updates or assigns values to global or workflow-specific variables. * Often paired with **Conditions** for dynamic updates. * Examples: * Set `escalation_required = true` if SLA is breached * Update `retry_count` after an API failure * Change `status = resolved` when conditions are met *** ### Delay The **Delay block** pauses the workflow for a set time before continuing. * Useful when external systems need time to update or respond. * Examples: * Wait **5 seconds** after an API call before fetching results * Add a **2-hour delay** before sending a follow-up reminder * Space out retries when contacting a third-party system Use **short delays** for API syncs and **longer delays** for reminders, follow-ups, or staged notifications. *** # Workflows Source: https://guides.robylon.ai/product/product Design reliable, debuggable AI-powered workflows using Triggers, Talk, Listen, Gen AI, Logic, Dev, and Notes blocks. This guide explains what each block does, when to use it, and shows end‑to‑end patterns. > Who is this for? Product managers, solutions engineers, and implementers building chat, voice, email, and ticketing automations. *** ## TL;DR * **Triggers** define **when** an automation starts (a chat opens, a voice call arrives, a ticket is created or updated, or another automation ends). * **Talk** sends outputs: chat messages, images, and **emails** (as internal notes, external notes, or customer replies) to the right audiences. * **Listen** collects inputs: button choices, text fields, images, and structured **captures** like order IDs or user info. * **Gen AI** powers reasoning: an **AI Agent** orchestrates tools; **Generate** composes answers; **Extract** pulls entities. * **Logic** gates flow with **Conditions**, **Set variables**, and **Delay** to sequence actions. * **Dev** connects systems via **API**, custom **Function** (Python), and **Integrations** (e.g., Google Sheets). * **Notes** improve readability and maintenance. *** ## Prerequisites * A project with at least one **channel** (Chat, Voice, or Email/Tickets) * Access to your **ticketing/CRM** sandbox and **API** credentials (if integrating) * Test users and **sample tickets/chats** *** ## Quickstart — Your First Resolution Flow Choose a **channel-specific trigger**: * **Chat**: On chat created / user message / handover from bot * **Voice**: On inbound call / scheduled outbound callback * **Email/Ticket**: On ticket created / reopened / updated / automation-ended Add **Capture** to parse essentials (e.g., `order_id`, `email`). Use **Buttons** for guided choices (status, returns, billing). Allow **Upload image** when proofs (POD, receipt, damage photos) are needed. Add one **AI Agent** per workflow. Bind tools: **Generate** for responses, **Extract** for entities, **KB** and **APIs** for facts. Use **Conditions** to separate happy vs. exception paths. **Set variables** for flags; **Delay** to wait for downstream systems. Call **API** to fetch/update records; use **Function** for validations (e.g., transaction hash). Read from **Integrations** like Sheets. Send a **Message** in chat or a structured **Email**: internal note for debug, external note for sandbox tests, or **reply** to the customer when live. Add **Notes** for maintainers. Emit debug **Talk → internal notes**. Close the loop or **Trigger** a follow‑on automation. Define when an automation starts or chains into another workflow. Send outputs like messages, emails, or debug notes to the right audience. Capture user inputs, intents, and data from systems or APIs. Power workflows with AI agents that generate, extract, and reason. Control workflow decisions with conditions, variables, and delays. Extend workflows with APIs, Python functions, integrations, and notes. ## Patterns ### 1) Ticket Created → Verify → Respond * **Trigger**: Ticket created * **Listen**: Capture `intent`, `order_id` * **Dev/API**: Fetch order * **Logic**: Branch on status; set variables * **Gen AI/Generate**: Compose resolution with policy guardrails * **Talk/Email**: Reply to ticket; log internal note ### 2) Chat Handover → Diagnose → Escalate * **Trigger**: Chat handed over from bot * **Listen**: Buttons for *billing*/*returns*/*technical* * **Gen AI/Extract**: Pull entities; **API** to CRM * **Logic**: If high‑value, escalate; else self‑serve path * **Talk/Message**: Provide next steps or connect human ### 3) Voice Inbound → Authenticate → Callback * **Trigger**: Inbound call * **Listen**: Capture account digits via DTMF * **Dev/Function**: Validate ID * **Logic/Delay**: Wait 5s for CRM lock release * **Talk**: Confirm and schedule **outbound** callback *** ## Production Checklist * One **AI Agent** per workflow * Clear tool contracts; consistent error shapes * PII handling: mask in logs/internal notes * Guardrails: KB only vs. model‑free‑form * Observability: internal notes + analytics events * Rollout: canary orgs, kill‑switch, monthly AI cap *** ## Troubleshooting Check **Email** mode. Use **Internal note** for debug; **External note** to post to the ticket without emailing the customer; **Reply to ticket** for customer‑facing. Add regex patterns and enable **Extract** fallback. Offer **Buttons** for guided flows when confidence is low. Add **Delay** after updates; implement retries with backoff; surface failures via **Internal note**. *** ## FAQs **How many AI Agents can I add?** One per workflow. Split flows if you need multiple agents. **When should I use External note vs Reply?** Use **External note** during UAT so staff can see drafts inside the ticketing tool. Switch to **Reply** when confident. **Can I start an automation from another automation?** Yes—use **automation‑ended** triggers to chain flows. *** ## Reference ### Block Matrix | Block | Purpose | Example Use | Notes | | --------------------- | ------------------- | ----------------------------------- | --------------------------- | | Trigger | Start flow on event | Ticket created; Chat handed over | Channel‑specific | | Talk → Message | Chat output | Status update; next steps | Use for customer replies | | Talk → Email | Ticket output | Internal note; External note; Reply | Choose visibility correctly | | Listen → Buttons | Input choices | Returns / Billing / Tech | Reduces ambiguity | | Listen → Capture | Entity extraction | order\_id, email, intent | Can pair with **Extract** | | Listen → Upload image | Collect proofs | Damage photos, receipts | Validate before action | | Gen AI → AI Agent | Orchestrator | Calls tools & KB | One per workflow | | Gen AI → Generate | Compose answer | Policy‑safe response | Add evaluation prompts | | Gen AI → Extract | Entities | Name, ID, amounts | Use for fallback parse | | Logic → Conditions | Branching | Eligibility, flags | Keep branches shallow | | Logic → Set variables | State mgmt | Set `is_vip` | Trace changes via notes | | Logic → Delay | Sequencing | Wait for sync | Keep short & purposeful | | Dev → API | System calls | CRM, OMS, Payments | Standardize headers/errors | | Dev → Function | Custom code | Validate hash | Python runtime | | Dev → Integration | 3rd‑party data | Google Sheets | Good for ops teams | | Notes | Maintainability | Explain rationale | Required in long flows | *** # Talk Components Source: https://guides.robylon.ai/product/talk Talk components are used to output information in workflows. They function like print statements in programming—controlling what is displayed, where it is displayed, and to whom. Messages can be made visible to: * **Only Robylon** (private debug/test) * **Robylon + Client** (internal collaboration) * **Robylon + Client + Customer** (live responses) They are essential for sending replies, setting up debug logs, or testing messages privately before going live. *** ### Message The **Message box** sends chat messages. * **Customer-visible replies** → Provide answers in real-time chats. * **Internal debug/test** → Validate how the workflow responds without exposing to customers. *** ### Image The **Image type** allows chatbots to **receive images from users** for analysis or processing.\ Typical uses: * Customer uploads **ID proof** for KYC. * User shares **screenshots** of errors. * Agents analyze **product photos** for returns. *** ### Email The **Email type** is used for sending structured responses or notes within the client’s ticketing system. It supports three modes: * **Internal Note** * Visible only to Robylon users. * Best for debugging messages, agent collaboration, or logging workflow steps. * **External Note** * Visible on the client’s ticketing system (e.g., Zendesk) but **not** visible to the customer. * Useful for **testing replies** before exposing them live. * **Reply to Ticket** * Sent directly to the **customer**. * Used when workflows are live and customer-ready. Start with **internal notes** for safe testing, then move to **external notes**, and finally **reply-to-ticket** when confident in workflow accuracy. *** # Triggers Source: https://guides.robylon.ai/product/triggers Triggers are the start points of an automation. They fire when a specific event occurs and launch the configured workflow. Use them to control when an automation begins and to chain automations when one ends. ### When to use * Start a workflow the moment a **customer action** or **system event** happens. * **Hand over** from one automation to another (e.g., “automation-ended”). * Keep flows **event-driven** and **composable** instead of building one mega-flow. ### Channel-specific trigger types | Channel | Common Events | Example Uses | | ------------------- | --------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | | **Chat** | **Chat created**, **User message**, **Bot handover**, **Automation ended** | Start when a **customer initiates chat**, or when a **previous bot** hands over to this flow | | **Voice** | **Inbound call received**, **Scheduled outbound call** | Launch IVR/auth flow when a **customer calls the helpline** or when the **system dials** | | **Email / Tickets** | **Ticket created**, **Ticket reopened**, **Ticket updated**, **Automation ended** | Fire on **new ticket**, **customer reply**, or **status change** to continue processing | ### Examples * **Ticket created → Auto-triage** and add tags; if missing info, **email** the user for details. * **Chat bot handover → Returns flow** with order lookup and RMA status. * **Inbound call → Verify → Callback** flow with scheduled follow-up if lines are busy. ### Chaining & Handover * Use **automation-ended** to start a **follow-on** automation after a flow completes. * Pass **context variables** (e.g., `intent`, `order_id`) to the next automation for continuity. ### Best practices * Keep triggers **small and focused**; one clear entry per flow. * Use **specific events** (e.g., “ticket reopened”) to avoid unintended runs. * **Log** why/when a trigger fired (internal notes) for observability. * Guard against re-entry with **idempotency keys** or **status checks** when applicable. ### Configure a trigger (Quick Steps) Pick **Chat**, **Voice**, or **Email/Tickets** based on where the event occurs. Choose the exact **event** (e.g., **Chat created**, **Ticket updated**, **Inbound call**). Capture key context (e.g., `user_id`, `ticket_id`, `source`) for downstream blocks. Route to your first block (**Listen**, **Gen AI**, or **Dev/API**) and set guardrails. Use **automation-ended** triggers to orchestrate multi-stage journeys (e.g., **Triage → Resolution → Feedback**), keeping each stage independently testable and debuggable.