Self-Hosted River

Self-Hosted River runs the entire River stack on your own infrastructure. No cloud dependency. No data crossing your network boundary. One license key. Available as a one-time purchase ($299) for individuals and teams.

Self-hosted is the right choice for:

• Healthcare, legal, government, financial-services teams with strict data residency
• Companies with significant on-premise compute (NVIDIA workstations, Mac clusters)
• Privacy-critical individuals
• Anyone who'd rather not pay per-user-per-month subscriptions

What's included#

The full local stack runs in Docker Compose:

• River backend (FastAPI, Python 3.11)
• WebSocket server for multiplayer sync
• Tool runtime for Agents and Workflows
• MongoDB for entity storage
• OpenSearch for full-text search
• MinIO for file storage (S3-compatible)
• Ollama for local AI inference (optional but recommended)

Everything runs on a single machine for personal use, or on a small cluster for a team.

What you control#

• The data — never leaves your machine
• The AI — point at any Ollama instance you choose
• The integrations — same connector framework, all running locally
• The updates — pull new versions on your schedule
• The compliance posture — match your environment's requirements exactly

Hardware#

For local AI specifically, the larger the model you can run, the better the agent loop performs.

Setup#

The standard setup is one Docker Compose file:

# Pull the self-hosted bundle
git clone https://github.com/river/self-hosted

# Configure your .env (paths, ports, license)
cp .env.example .env

# Start the stack
docker compose up -d

# Visit https://your-host

Detailed instructions, system requirements, and configuration options are in the [Self-Hosted Deployment Guide](/docs/Self Hosted River/).

License & offline grace#

Self-Hosted is licensed per workspace. You purchase once, receive a license key, and the stack runs indefinitely. There's a 7-day offline grace period for license validation, so a brief connectivity issue doesn't disrupt work.

Pair with local AI#

Self-Hosted is most powerful when paired with local Ollama models. The combination gives you:

• Zero per-token AI cost
• Full control over which models run
• Inference that never leaves your machine
• The same River feature set as cloud

For teams with serious GPU resources, this can replace tens of thousands of dollars in cloud AI spend.

What still touches the cloud (optional)#

Even self-hosted, some things may use external services if you configure them:

• Connected integrations (Gmail, Stripe, etc.) call those services' APIs as you'd expect
• The Browser Extension communicates with your self-hosted backend via your network
• Cloud AI providers (OpenAI, Anthropic, etc.) only used if you explicitly configure them

You can run a fully air-gapped configuration with no external services at all if your environment requires it.

Tips#

• Start with Mac Studio or a single GPU workstation for evaluation. Scale up only after you've validated the workflow.
• For sensitive Dataspaces, run them in Self-Hosted; keep less-sensitive personal use on River Cloud. You can be in both at the same time.
• The Self-Hosted bundle includes monitoring dashboards. Use them — they catch issues earlier than user reports do.