MATE — Command Center for AI Agents

What does this agent do, and when should you use it?

MATE is a production-oriented multi-agent orchestration engine built on Google ADK (with LangGraph support), served through FastAPI with a web dashboard. Agent definitions live in the database and are edited on a drag-and-drop canvas where you draw parent→child hierarchies, attach tools and MCP servers, and roll back to any previous version in one click. It reaches 50+ LLM providers through LiteLLM — including local servers such as Ollama and LM Studio — and logs every model call with a US-dollar cost plus a four-bucket token breakdown. Governance features include per-agent RBAC, token and dollar budgets that warn or block, alert rules to webhook/email/Slack/Discord, guardrails, and an append-only audit log. An eval framework with exact-match, semantic-similarity and LLM-judge scoring, plus a Playground for side-by-side variant comparison, covers pre-release testing. It ships as a Docker image or run-from-source self-hosted service with SQLite by default and PostgreSQL/MySQL 8 as alternatives.

MATE boots a FastAPI server via auth_server.py on port 8000 and reads agent definitions from the database: model, instructions, parent/child hierarchy, tools, MCP servers, memory blocks, planner and RBAC roles. During a conversation it calls the configured model through LiteLLM and writes each call's dollar cost and prompt/response/thoughts/tool-use token counts to logs surfaced at /dashboard/usage. It exposes each root agent as an MCP server (/agents/{name}/mcp/*) and an A2A endpoint (/a2a/{agent}), provides an OpenAI-compatible /v1 bridge, and ships an embeddable <script> chat widget. The eval framework invokes the live agent, scores responses by exact match, semantic similarity or LLM-as-Judge, and fires a regression webhook when a new version drops more than 5 points. Database migrations run automatically at startup.

1. Platform or ops teams that need one control plane for agent hierarchies, permissions and costs across multiple teams and projects.
2. Product teams that iterate on prompts and models without redeploying, with one-click rollback as a safety net.
3. Teams deploying chat on several public websites (support, sales, docs) with per-site appearance and access rules via widget keys.
4. Finance or compliance groups tracking LLM spend per agent, project and user, with budgets and alerting.
5. Developers connecting agents to Claude Desktop, Cursor or OpenAI-compatible IDE clients.
6. Teams wanting automated regression suites against prompt changes before shipping.

How do you install or deploy this agent?

Fastest path: run the published Docker image.

docker run -d --name mate -p 8000:8000 \
  -e AUTH_PASSWORD=change-me -e GOOGLE_API_KEY=your-key \
  -e DB_PATH=data/my_agent_data.db -v mate-data:/app/data \
  ghcr.io/antiv/mate:latest

Or install from source:

git clone https://github.com/antiv/mate.git && cd mate
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.lock

cp .env.example .env
# Set GOOGLE_API_KEY (or any supported provider key), AUTH_USERNAME and AUTH_PASSWORD

You can also build the full stack with docker-compose up --build (it needs AUTH_PASSWORD set in .env).

Requirements: Python 3.10+ for the source path, one LLM provider API key, and a database — SQLite by default, with migrations applied automatically on startup.

How do you use this agent?

After the service starts, open http://localhost:8000 and log in with AUTH_USERNAME / AUTH_PASSWORD (left empty they are admin / mate; MATE_ENV=production refuses that default).

Source install launch:

python auth_server.py
# Dashboard: http://localhost:8000

You land on /dashboard/workroom, pick a root agent from the card grid and chat. On /dashboard/agents you create agents, draw parent→child edges and configure tools and MCP servers on the canvas.

Expose agents as MCP servers:

export MCP_EXPOSED_AGENTS=creative_agent,support_agent

Clients authenticate with a Personal Access Token from the dashboard account menu:

curl -H "Authorization: Bearer mate_pat_..." https://your-mate-instance.com/mcp

Embed the chat widget on a website:

<script
  src="https://your-mate-instance.com/widget/mate-widget.js"
  data-key="wk_abc123..."
  data-server="https://your-mate-instance.com"
></script>

Common environment variables: DB_TYPE, GOOGLE_CLIENT_ID/GOOGLE_CLIENT_SECRET and GITHUB_CLIENT_ID/GITHUB_CLIENT_SECRET for SSO, RATE_LIMIT_ENABLED, ALERTS_ENABLED, CSP_MODE, OTEL_TRACING_ENABLED, AUDIT_RETENTION_DAYS.

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ