Agents

Agents are the core of auxilia. An agent is an AI assistant configured with a system prompt (instructions), an emoji avatar, and a set of MCP server bindings that determine which tools it can use.

How Agents Work

When a user sends a message to an agent, auxilia:

1. Loads the agent configuration (instructions, model, MCP servers)
2. Connects to all bound MCP servers and discovers available tools
3. Filters tools based on each tool’s status (always allow / needs approval / disabled)
4. Sends the message to the LLM with the system prompt and available tools
5. Streams the response back, executing tool calls as needed

Creating an Agent

1. Navigate to the Agents page
2. Click Create Agent
3. Fill in the agent details:
   • Name — a descriptive name for the agent
   • Emoji — an avatar emoji shown in the agent list and chat
   • Instructions — the system prompt that defines the agent’s behavior

Agent-MCP Server Bindings

Each agent can be connected to multiple MCP servers. When you bind an MCP server to an agent, all of that server’s tools become available (subject to tool settings).

To add MCP servers to an agent:

1. Open the agent’s configuration page
2. Click Add MCP Server
3. Select from the workspace’s registered MCP servers
4. Configure tool permissions for each server

Chat Threads

Conversations with agents are organized into threads. Each thread:

• Is tied to a specific agent and user
• Has its own conversation history (persisted via LangGraph checkpoints)
• Can use a specific LLM model (selectable per thread)
• Is independent from other threads with the same agent

Supported Models

You can select a different LLM model for each thread. Available models depend on which API keys are configured: