psyb0t/chatz repository overview

Source⁠

⁠chatz

Self-hosted AI chat with OpenAI-compatible and Anthropic models, MCP tools, and an embedded web app. Run one Docker image with SQLite, or use Postgres when needed.

⁠Features

• Streaming conversations. Messages, reasoning, and tool calls arrive as the model produces them. Stop or refresh without losing the submitted prompt.
• Models you already use. Connect OpenAI-compatible or Anthropic endpoints. Chatz discovers their models and keeps provider-specific settings in the chat.
• MCP tools. Admins add stdio or HTTP MCP servers, then users enable their tools per chat. Stored MCP settings are encrypted. Authorization headers stay masked in admin responses.
• Useful chat controls. Set temperature, reasoning, output size, and the history budget per chat. Create, rename, search, and delete chats.
• Generative UI. Model and tool output can render tables, status panels, charts, log viewers, and other live components in the conversation. See the rendering guide⁠ for the component catalog.
• Admin-managed access. The first user is the admin. That admin creates the other accounts and configures upstreams and MCP servers.
• Demo mode. Set CHATZ_SHOWCASE_MODE=true to return fixed streamed replies for exact demo prompts. Normal chats keep using the selected model.

⁠Quickstart

Chatz ships as the psyb0t/chatz Docker image. The fastest path uses SQLite in a persistent Docker volume. It needs no LLM configuration for the first run.

docker pull psyb0t/chatz:latest
docker volume create chatz-data
touch chatz.log
sudo chown 1000:1000 chatz.log
docker run --name chatz --detach --restart unless-stopped --init \
  --publish 127.0.0.1:8080:8080 \
  --env CHATZ_DB_DRIVER=sqlite \
  --read-only \
  --tmpfs /tmp:rw,noexec,nosuid,size=64m \
  --cap-drop ALL \
  --security-opt no-new-privileges:true \
  --memory 512m \
  --cpus 1.0 \
  --pids-limit 256 \
  --mount type=bind,src="$(pwd)/chatz.log",dst=/app/chatz.log \
  --volume chatz-data:/data \
  psyb0t/chatz:latest run

Open http://localhost:8080⁠, then visit /setup to create the admin. The chat works before you configure an LLM, but the model picker stays empty.

latest is for a first look. Pin an immutable release tag such as psyb0t/chatz:v0.7.9 for deployment. Copy .env.example to .env, set CHATZ_DB_DRIVER=sqlite, and add --env-file .env to configure upstreams, MCP secret storage, or other settings. Chatz exposes GET /healthz.

⁠Try fixed demo replies

Start the normal stack with fixed-response mode:

CHATZ_SHOWCASE_MODE=true docker compose up --build

After completing /setup, select a configured model and send, for example:

• Show me what's happening across the production platform right now.
• Where are we losing deals in the sales pipeline?
• Which customers are at risk and who should the team contact first?

Each returns a paced reply with thinking, tool calls, and a dashboard. Any other message uses the selected model and normal MCP-enabled chat path.

⁠Connect a model

Copy .env.example to .env and set CHATZ_UPSTREAMS. Each upstream chooses the openai driver, which also works with OpenAI-compatible APIs such as AIGate⁠, or the anthropic driver. apiKeyEnv names a variable in .env; never put the key in the JSON.

[
  {"name":"openai","provider":"openai","baseUrl":"https://api.openai.com/v1","apiKeyEnv":"OPENAI_API_KEY"},
  {"name":"anthropic","provider":"anthropic","apiKeyEnv":"ANTHROPIC_API_KEY"},
  {"name":"aigate","provider":"openai","baseUrl":"http://aigate:4000","apiKeyEnv":"AIGATE_TOKEN"},
  {"name":"ollama","provider":"openai","baseUrl":"http://localhost:11434/v1"}
]

Docker Compose loads .env automatically. Add --env-file .env to the Docker command above. The AIGate example assumes both containers share a Docker network. Use AIGate's reachable root URL otherwise. The example environment file⁠ documents model metadata, timeouts, and every other setting.

Generate a 32-byte base64 key before adding MCP credentials, then set it as CHATZ_SECRETS_KEY:

openssl rand -base64 32

Without this key, Chatz refuses to store MCP HTTP headers or stdio environment variables.

⁠Configuration

All configuration comes from environment variables. Start with .env.example⁠. The usual settings are:

• CHATZ_DB_DRIVER=sqlite for one local container, or postgres for Compose and shared deployments.
• CHATZ_UPSTREAMS and the key variables named by apiKeyEnv for models.
• CHATZ_SECRETS_KEY before storing MCP credentials. Set CHATZ_AUTH_PASSWORDLESS=true for a single-user installation.
• LOG_LEVEL=debug for local prompt troubleshooting. Debug logs may contain ordinary chat text, so do not send them to a shared log service.

⁠Architecture

The Go service serves the API and embedded web app, stores chat data in Postgres or SQLite, calls models through Elelem, and connects MCP servers on demand. See the architecture guide⁠ for the details.

⁠HTTP API

The versioned JSON and SSE API is under /api/v1. Read api/api.yml⁠ for the contract. /healthz is available for container health checks.

⁠Deploy

Every push to main publishes psyb0t/chatz:latest. A v<semver> release tag publishes the matching immutable psyb0t/chatz:v<semver> image. Deploy a tag, not latest. Use SQLite for one local volume and Postgres for replicas, networked storage, or an existing Postgres installation. For Docker Compose and development, read getting started⁠ and development⁠.

⁠License

See LICENSE⁠.