Configure an Agent

What is an agent?

An agent is a configurable AI endpoint on top of one or more deployments. It gives you:

• a stable slug used as the model field in API requests
• a routing strategy — Fixed, Rule-Based, or Workflow
• a system prompt prepended to every request
• optional Knowledge Bases and MCP Tools

To create an agent you need at least one model deployed. If you haven't done that yet: Create a dedicated deployment.

---

Create an Agent

Navigate to Agents in the left sidebar and click New Agent in the top-right corner.

In the creation dialog you can set:

• Name — a descriptive name for your agent
• Description — an optional short description
• Routing Strategy — the initial strategy (can be changed at any time)

Mycelis supports three routing strategies:

Strategy	Description
Fixed	A single fixed deployment handles all requests
Rule-Based	Rules determine which deployment receives a request
Workflow	Visual graph editor for complex pipelines

Click Create to open the agent's detail view.

---

General Tab

The General tab lets you configure the basics:

• Name — rename the agent
• Slug (API Model ID) — the unique identifier used as the model field in API requests. Important for OpenAI-compatible clients like OpenCode, Cursor, and others
• System Prompt — a system-level instruction prepended to every request sent through this agent

---

Models & Routing Tab

This is where you configure or switch the routing strategy.

Routing Strategy: Fixed

Every request is forwarded to a single, statically configured deployment.

1. Select Fixed as the strategy
2. Pick the target deployment from the dropdown
3. Click Save

After saving, two additional tabs appear: Knowledge & Tools and Settings.

Routing Strategy: Rule-Based

Rules determine which deployment handles a request — based on conditions like the detected intent, query content, or technical request properties.

Click New Rule to create a rule. Each rule consists of one or more Conditions and a target deployment.

Condition: Auto Intent

Mycelis automatically detects request intent. Available intents:

• Coding, Debugging, Code Review, Architecture, Explanation, Documentation
• Creative Writing, Data Analysis, Math, Translation, DevOps, Agentic, Long Context

Condition: Text / Query

• Contains — the request contains a specific word or phrase
• Regex — the request matches a regular expression

Condition: Request Info

• Has Tools — the request includes tool definitions
• Tokens > N / Tokens < N — token count threshold
• Client Type Is — Cursor, GitHub Copilot, Continue, OpenCode, Generic
• MCP Tool Available — a specific MCP tool is present in the request
• Knowledge Base ID — a specific Knowledge Base is active

Add a Default Rule (fallback) to ensure every request gets a response when no other rule matches.

Routing Strategy: Workflow

The Workflow editor is a visual graph editor for building complex routing pipelines. Knowledge Bases and Tools are configured directly inside the editor.

Start from one of the available templates or begin with a blank canvas.

Available node types:

Node	Description
Start	Entry point — every pipeline begins here
Raw Model	Forwards the request directly to a deployment
Agent	Forwards the request to another configured agent
If / Else	Branching node with configurable If and Else branches
Knowledge Base	Attaches a single Knowledge Base as context
All KBs	Attaches all Knowledge Bases in the workspace
MCP Tool	Calls a single MCP tool
MCP Server	Attaches a complete MCP server
MCP Hub	Makes all activated MCP tools available
Semantic Cache	Caches semantically similar requests for faster responses
Smart Router	Intelligent routing based on request properties

Connect the nodes to define the desired data flow, then click Save.

---

Knowledge & Tools Tab

Available for Fixed and Rule-Based agents. In Workflow mode, configure Knowledge Bases and Tools directly in the editor.

MCP Tools

Add entire MCP servers or selectively enable individual tools within a server.

If you haven't set up MCP tools yet, do so first in the MCP Hub.

Knowledge Bases

Attach one or more Knowledge Bases. For every request, the agent automatically retrieves relevant passages and adds them as context.

Haven't created a Knowledge Base yet? Learn how to set up a RAG pipeline.

---

Call via API

Use the agent's slug as the model field:

curl https://app.mycelis.ai/api/proxy/v1/chat/completions \
  -H "Authorization: Bearer pat_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-agent-slug",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

---

Next steps

• Set Up a RAG Pipeline — attach documents as a knowledge source
• MCP Tools — use external APIs directly in your model context
• API Reference — all available API parameters