DocsAgents
Agents are AI assistants defined within blueprints. They have configurable identities, knowledge bases, and flows that power conversations and automated interactions.
An agent is an AI assistant defined within a blueprint. Each agent has an identity (name, avatar, description), a behavior model powered by flows and LLM interactions, and optionally a knowledge base for retrieval-augmented generation.
Agents are the primary way users interact with AI on the platform — through conversations, web widgets, or messaging channels.
How agents work
When a user sends a message to an agent, the following happens:
- The message is posted to a conversation where the agent is a participant.
- If the agent has
autoRespondenabled, its default flow is triggered. - The flow executes its steps — typically calling the assistant engine, which:
- Retrieves relevant documents from the agent's knowledge base
- Builds a prompt with conversation context and retrieved content
- Calls the configured LLM to generate a response
- Streams the response back to the conversation
- The user sees the agent's response in real time.
User message
→ Conversation
→ Agent's default flow triggered
→ Assistant step
→ Knowledge base retrieval (RAG)
→ LLM generation
→ Response streamed to conversationAgent properties
| Property | Description |
|---|---|
| Name | Display name shown to users in conversations |
| Description | Context about the agent's role — used as system instructions for the LLM |
| Avatar | Visual identity (image) |
| Color | Accent color for the agent in the UI |
| Default flow | The flow triggered when the agent participates in conversations |
| Knowledge base | A search index used for RAG retrieval |
| Query suggestions | Pre-written prompts suggested to users at the start of a conversation |
| Service agent | Whether this is a background agent (not user-facing) |
Agent identity
Each agent has its own identity on the platform. When an app is installed, the platform creates a service account for each agent in the identity provider. This gives agents:
- A persistent identity across conversations
- The ability to post messages as themselves
- Proper attribution in conversation history
- Permission to execute operations on behalf of the app
In this section
- Configuring Agents — creating and setting up agents within blueprints
- Knowledge Base — building a knowledge base for RAG-powered responses
- Assistant Execution — how the AI engine processes queries and generates responses
Build-time vs. Runtime Data
The Bivariant platform separates build-time definitions (blueprints) from runtime data (per-organization). This separation is fundamental to multi-tenant isolation and safe distribution.
Configuring Agents
How to create and configure agents within a blueprint — setting identity, behavior, knowledge bases, and default flows.