Skip to main content
The chat interface is where you and your users interact with agents. It supports real-time streaming, file attachments, knowledge base citations, and full conversation history.

​
Starting a Conversation

To chat with an agent:
  1. Navigate to your agent’s page.
  2. Click Chat to open the chat interface.
  3. If the agent has a welcome message configured, it appears automatically as the first message.
  4. Type your message and press Enter or click Send.
Each conversation is a separate session with its own context and memory. The agent remembers everything said within a conversation but starts fresh in a new one.

​
Streaming Responses

When streaming is enabled, you see the agent’s response appear word-by-word in real time rather than waiting for the full response to generate. This makes conversations feel more natural and responsive. Streaming is active when:
  • The agent has streaming_enabled turned on (default: yes)
  • The client supports streaming (the web interface always does)
If your agent is deployed through channels like Telegram or WhatsApp, responses are delivered in full once generation completes, since those platforms do not support real-time streaming.

​
Citations and Sources

When your agent has linked knowledge bases, responses may include numbered citations like [1], [2], [3]. These indicate that the agent used specific content from your knowledge base to generate its answer.

​
Citation Tooltips

Hover over any citation number in a response to see a tooltip with:
  • Knowledge base name — Which KB the source came from
  • Source document — The filename or URL of the original document
  • Relevance score — How closely the content matched the query (0-100%)
  • Chunk text — A preview of the actual content that was retrieved

​
Sources Panel

Click the Sources button (or a citation) to open the Sources Panel — a slide-in sidebar that shows all sources used in the current response:
  • Sources are grouped by knowledge base
  • Each source card is expandable to show the full chunk text
  • Relevance scores are displayed as color-coded bars:
    • Green (high relevance) — Strong match
    • Yellow (medium) — Moderate match
    • Red (low) — Weaker match but still above threshold
  • Platform badges show the original source type (PDF, URL, Instagram, etc.)
  • View Original links take you to the source document or URL when available
The sources panel is the best way to verify your agent’s responses. If a citation does not support the claim, you may need to adjust your knowledge base content or system prompt.

​
Conversation Memory

Each conversation maintains context through LangChain conversation memory:
  • The agent remembers all previous messages in the current conversation
  • Context accumulates over the conversation, allowing follow-up questions
  • Knowledge base search uses the last 5 messages as context for better retrieval
Memory is conversation-scoped — starting a new conversation resets the context.

​
Dynamic Variables in Chat

If your agent uses dynamic variables ({{variable_name}}), you can pass them through:
  • Query parameters when linking to the chat (using var_ prefix)
  • Widget configuration when embedding the agent
  • API calls when using the chat endpoint programmatically
Variables can also be updated mid-conversation by including a variables field in subsequent messages. This enables dynamic experiences where context changes as the conversation progresses (e.g., updating the current page the user is on).

​
Public Agent Pages

Published agents are accessible at /a/your-agent-slug. These pages:
  • Require no authentication
  • Show the agent’s name and description
  • Provide a full chat interface
  • Support all features including citations, streaming, and attachments

​
Chat via API

You can also interact with agents programmatically:
The API returns the same response structure as the web interface, including the message text and any sources/citations.