Skip to content

Getting Started

This guide will help you set up MindRoom and create your first AI agent.

If you do not want to self-host Matrix yet, this is the simplest setup. You only run MindRoom locally. Watch the 2-minute setup video:

MindRoom: installing and talking to my first AI agent in 2 minutes

Prerequisite: Install uv.

1. Initialize local config

uvx mindroom config init

This creates:

  • ~/.mindroom/config.yaml
  • ~/.mindroom/.env prefilled with MATRIX_HOMESERVER=https://mindroom.chat

The default --matrix-server mindroom.chat preset uses hosted Matrix and defaults to the openai provider. Use --provider to select a different provider preset:

# Use Anthropic Claude
uvx mindroom config init --provider anthropic

# Use Azure OpenAI
uvx mindroom config init --provider azure

# Use a Codex CLI ChatGPT login
uvx mindroom config init --provider codex

# Use local Ollama
uvx mindroom config init --provider ollama

# Use local llama.cpp through its OpenAI-compatible server
uvx mindroom config init --provider llama.cpp

# Use Vertex AI Claude (Google Cloud)
uvx mindroom config init --provider vertexai_claude

Use --matrix-server mindroom.chat for hosted Matrix and --matrix-server self-hosted when you run your own homeserver. Use --provider for the model provider. Run codex login before starting MindRoom when using --provider codex.

--provider azure uses Azure OpenAI through your deployment name. Set models.default.id to the Azure deployment name you created.

--provider ollama uses local Ollama with gemma4 by default and also configures qwen3.6:27b. Run ollama pull gemma4 and ollama pull qwen3.6:27b before starting MindRoom.

--provider llama.cpp uses a local OpenAI-compatible llama.cpp server on http://localhost:8080/v1. Start llama-server with one of the configured Unsloth GGUF refs before starting MindRoom. These local provider configs run entirely locally and do not require real cloud API keys such as OPENAI_API_KEY or ANTHROPIC_API_KEY unless you switch the config to a remote provider.

Use --provider vertexai_claude for Vertex AI Claude on hosted Matrix.

2. Add remote-provider credentials when needed

$EDITOR ~/.mindroom/.env

For hosted providers, set the credentials for the provider you selected:

  • ANTHROPIC_API_KEY=..., or
  • AZURE_OPENAI_API_KEY=... and AZURE_OPENAI_ENDPOINT=..., or
  • OPENAI_API_KEY=..., or
  • OPENROUTER_API_KEY=..., or
  • For Codex CLI ChatGPT authentication: run codex login.
  • For Kimi Code CLI authentication: run kimi and /login.
  • For Vertex AI Claude: set ANTHROPIC_VERTEX_PROJECT_ID and CLOUD_ML_REGION and authenticate with gcloud auth application-default login. Skip this step for --provider ollama or --provider llama.cpp unless you also add a remote provider.

3. Pair your local install from chat UI

  1. Open https://chat.mindroom.chat and sign in.
  2. Go to Settings -> Local MindRoom.
  3. Click Generate Pair Code.
  4. Run locally:
uvx mindroom connect --pair-code ABCD-EFGH

Notes:

  • Pair code is short-lived (10 minutes). Generate a new one if it expires.
  • mindroom connect writes local provisioning values (including MINDROOM_NAMESPACE) into ~/.mindroom/.env by default.
  • Use --no-persist-env to print export commands instead of writing .env; evaluate or copy those commands into the current shell yourself.

4. Run MindRoom

uvx mindroom run

5. Verify

In chat: Send a message mentioning your agent in a room where it is configured.

Dashboard: Access the web dashboard at http://localhost:8765 to configure agents, models, and tools. Protect the dashboard API in non-localhost environments by setting MINDROOM_API_KEY in your .env.

Preflight check: Run uvx mindroom doctor before uvx mindroom run to verify config, API keys, Matrix connectivity, and storage in one pass.

For a detailed architecture and credential model, see: Hosted Matrix deployment guide.

Preferred alternative: NixOS LXC container (agent-controlled machine)

Use this when you want to give a MindRoom agent full freedom over its own virtual machine while you, from the host, control precisely what it can see. The mindroom-ai/lxc-nixos flake provisions the virtual machine — an Incus LXC system container running NixOS — with the full MindRoom stack (MindRoom, Tuwunel Matrix homeserver, MindRoom Chat, and Caddy) plus Docker and ragenix-based secrets wiring. Because the whole virtual machine is declared in the flake, the agent can rebuild and manage the persistent system it runs on — unlike the mostly stateless Docker Compose stack below — without ever touching the host. It is slightly harder to set up by hand, but asking a coding agent such as Codex or Claude Code to do it is trivial: the repo ships machine-oriented instructions in AGENTS.md. It requires a Linux host running Incus.

git clone https://github.com/mindroom-ai/lxc-nixos.git
cd lxc-nixos
incus launch images:nixos/unstable mindroom -c security.nesting=true
incus config device add mindroom repo disk source="$PWD" path=/mnt/repo shift=true

Then follow the repo README for operator SSH keys, secrets bootstrap, and the nixos-rebuild switch deployment flow.

Alternative: Full Stack Docker Compose (bundled dashboard + Matrix + MindRoom client)

Use this when you want everything local: the bundled MindRoom dashboard, Matrix homeserver, and a Matrix client in one stack.

Prereqs: Docker + Docker Compose.

1. Clone the full stack repo

git clone https://github.com/mindroom-ai/mindroom-stack
cd mindroom-stack

2. Add your API keys

cp .env.example .env
$EDITOR .env  # add at least one AI provider key

3. Start everything

docker compose up -d

Open:

  • MindRoom UI: http://localhost:8765
  • MindRoom client: http://localhost:8080
  • Matrix homeserver: http://localhost:8008

The stack uses published mindroom, mindroom-chat, and mindroom-tuwunel images by default.

If you access the stack from another device, set CLIENT_HOMESERVER_URL=http://<host-ip>:8008 in .env before starting it.

Manual Install (advanced)

Use this if you already have a Matrix homeserver and want to run MindRoom directly.

Prerequisites

  • Python 3.12 or higher
  • A Matrix homeserver (or use a public one like matrix.org)
  • API keys for your preferred AI provider (Anthropic, OpenAI, etc.)

Installation

uv tool install mindroom

pip install mindroom

git clone https://github.com/mindroom-ai/mindroom
cd mindroom
uv sync
source .venv/bin/activate

Configuration

1. Create your config file

Create a config.yaml in your working directory:

agents:
  assistant:
    display_name: Assistant
    role: A helpful AI assistant that can answer questions
    model: default
    include_default_tools: true
    rooms: [lobby]
    # Optional: file-based context (OpenClaw-style)
    # context_files: [SOUL.md, USER.md]

models:
  default:
    provider: openai
    id: gpt-5.6

defaults:
  tools: [scheduler]
  markdown: true

authorization:
  global_users:
    - "@alice:matrix.example.com"
  agent_reply_permissions:
    "*":
      - "@alice:matrix.example.com"

timezone: America/Los_Angeles

2. Set up environment variables

Create a .env file with your credentials:

# Matrix homeserver must allow managed account provisioning.
MATRIX_HOMESERVER=https://matrix.example.com

# Recommended when the homeserver supports token-gated registration.
# MATRIX_REGISTRATION_TOKEN=your-registration-token

# Optional: For self-signed certificates (development)
# MATRIX_SSL_VERIFY=false

# Optional: For federation setups where server_name differs from homeserver hostname
# MATRIX_SERVER_NAME=example.com

# AI provider API keys
OPENAI_API_KEY=your_openai_key
# OPENROUTER_API_KEY=your_openrouter_key
# ANTHROPIC_API_KEY=your_anthropic_key

# Optional: protect the dashboard API (recommended for non-localhost)
# MINDROOM_API_KEY=your-secret-key

Optional: Bootstrap local Synapse + MindRoom Chat with Docker (Linux/macOS)

If you want a local Matrix + client setup without running the full mindroom-stack app, use the helper command:

mindroom local-stack-setup --synapse-dir /path/to/mindroom-stack/local/matrix

If you're running from source in this repo, use:

uv run mindroom local-stack-setup --synapse-dir /path/to/mindroom-stack/local/matrix

This starts Synapse from the mindroom-stack compose files, starts a MindRoom Chat container, waits for both services to be healthy, and by default writes local Matrix settings to .env next to your active config.yaml.

Note

MindRoom automatically creates Matrix user accounts for each agent. Configure one supported provisioning path: hosted provisioning, MATRIX_REGISTRATION_TOKEN, MATRIX_REGISTRATION_SHARED_SECRET, or intentionally open registration. If registration fails, check the selected provisioning credentials and homeserver policy.

3. Run MindRoom

mindroom run

MindRoom will:

  1. Connect to your Matrix homeserver
  2. Create Matrix users for each agent
  3. Create any rooms that don't exist and join them
  4. Start listening for messages

Next Steps