Installation

Docker Compose

Docker Compose runs the platform and web UI as two containers. This is the recommended method for single-machine deployments.

1. Create a project directory

   mkdir nimblebrain && cd nimblebrain

2. Create nimblebrain.json

   nimblebrain.json

   {
     "$schema": "https://schemas.nimblebrain.ai/v1/nimblebrain-config.schema.json",
     "version": "1"
   }

   This is runtime config only (model, limits, features). Installed apps are not listed here — per-workspace app installs live in workspace.json, which the runtime manages. See the configuration reference for all options.

3. Create docker-compose.yml

   docker-compose.yml

   services:
     platform:
       image: ghcr.io/nimblebraininc/nimblebrain:latest
       restart: unless-stopped
       volumes:
         - workspace:/data
         - ./nimblebrain.json:/data/nimblebrain.json:ro
         - ./instance.json:/data/instance.json:ro
       environment:
         - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
         - OPENAI_API_KEY=${OPENAI_API_KEY:-}
         - GOOGLE_GENERATIVE_AI_API_KEY=${GOOGLE_GENERATIVE_AI_API_KEY:-}
         - NEBIUS_API_KEY=${NEBIUS_API_KEY:-}
         - XAI_API_KEY=${XAI_API_KEY:-}
       networks:
         - internal
   
     web:
       image: ghcr.io/nimblebraininc/nimblebrain-web:latest
       restart: unless-stopped
       ports:
         - "27246:8080"
       environment:
         - PLATFORM_URL=http://platform:27247
       depends_on:
         platform:
           condition: service_healthy
       networks:
         - internal
   
   volumes:
     workspace:
   
   networks:
     internal:

   The platform image runs on Bun. Connectors are remote MCP servers, so nothing you install runs inside this container. The runtime working directory inside the container is /data (set via NB_WORK_DIR), so the config mount lands at /data/nimblebrain.json.

4. Start the services

   export ANTHROPIC_API_KEY=sk-ant-api03-...
   docker compose up

   The platform API runs on port 27247 (internal). The web UI is published on port 27246. Health is checked internally at http://localhost:27247/v1/health.

   The platform does not start without instance.json (see Authentication below). For a local trial, write the dev adapter first: echo '{"auth":{"adapter":"dev"}}' > instance.json.

Build from local source

To build the images from a checkout instead of pulling :latest, add a build: key and use docker compose up --build:

docker-compose.yml (build from source)

services:
  platform:
    image: ghcr.io/nimblebraininc/nimblebrain:latest
    build: .
    # ...
  web:
    image: ghcr.io/nimblebraininc/nimblebrain-web:latest
    build: web/
    # ...

docker compose up --build

Development (Bun)

Run from source for development. Requires Bun 1.2.0 or newer (engines.bun is >=1.2.0).

1. Clone and install

   git clone https://github.com/NimbleBrainInc/nimblebrain.git
   cd nimblebrain
   bun install

2. Install web dependencies

   cd web && bun install && cd ..

3. Set your API key

   export ANTHROPIC_API_KEY=sk-ant-api03-...

4. Start development mode

   bun run dev

   This starts both the API server (with file watching and auto-restart) and the web client (Vite HMR) in a single terminal. The dev workdir is project-local (.nimblebrain/). The launcher writes the dev adapter to the workdir’s instance.json when it has none, so no auth setup is required locally.

Other dev commands:

bun run dev:empty     # Empty state — no apps, fresh workdir (:27249/:27248)
bun run dev:minimal   # Minimal — no apps, preset model config (:27251/:27250)
bun run test          # Run unit AND integration tests
bun run eval          # LLM evals (requires ANTHROPIC_API_KEY)

Environment variables

Variable	Required	Description
ANTHROPIC_API_KEY	Yes	Anthropic API key for Claude. At least one provider key is required.
OPENAI_API_KEY	No	OpenAI provider key.
GOOGLE_GENERATIVE_AI_API_KEY	No	Google provider key.
PORT	No	HTTP port for the API server. Default: 27247.
NB_PUBLIC_ORIGIN	No	Canonical public origin used to build absolute URLs (OAuth callbacks, post-login landing). Default in dev: http://localhost:27247. Set this in production. See Environment Variables.
NB_WORK_DIR	No	Working directory for conversations, skills, and app state. Default: .nimblebrain/ (project-local) or ~/.nimblebrain (global serve). Set to /data in the Docker image.
ALLOWED_ORIGINS	No	Comma-separated allowed CORS origins. Default: all origins.
NB_TELEMETRY_DISABLED	No	Set to 1 to disable anonymous telemetry.
DO_NOT_TRACK	No	Set to 1 to disable anonymous telemetry.

Note

There is no NB_API_KEY request-auth variable. Authentication is configured through instance.json adapters, not an environment-variable bearer key.

Authentication

<workDir>/instance.json names the identity provider, and the platform does not start without it:

• "adapter": "dev" — no login: every request resolves to a single built-in user. Fine for local development; not for shared or networked deployments.
• "adapter": "oidc" or "workos" — a real identity provider (any OpenID Connect issuer, or WorkOS). On successful login the provider’s callback sets an nb_session cookie.

instance.json (OIDC example)

{
  "auth": {
    "adapter": "oidc",
    "issuer": "https://your-idp.example.com",
    "clientId": "your-client-id",
    "allowedDomains": ["example.com"]
  }
}

Minimal config file

NimbleBrain reads nimblebrain.json at startup. A minimal config:

nimblebrain.json

{
  "$schema": "https://schemas.nimblebrain.ai/v1/nimblebrain-config.schema.json",
  "version": "1"
}

See the configuration reference for all options.

What’s next

Core ConceptsAgent loop, connectors, skills, tools, conversations

nimblebrain.json ReferenceFull configuration file reference

Docker DeploymentProduction Docker Compose configuration