OllaBridge ⚡️

Bring your own LLMs, devices, GPUs, and provider accounts. Use them securely from anywhere through one OpenAI-compatible API.

Quick Start • The Dashboard • Why OllaBridge • Local vs Cloud vs Enterprise • Docs

The OllaBridge control tower: your models, devices, providers, and cloud relay — live, in one place.

🎯 What is OllaBridge?

One gateway. All your LLMs. Everywhere.

OllaBridge turns your PC into a private, OpenAI-compatible LLM provider in ~60 seconds — and grows with you into a hybrid control plane for every model you can reach: local Ollama, remote GPUs, your team's workstations, and your own OpenAI/Anthropic/Gemini accounts.

The problem: your models live everywhere (laptop, cloud GPU, gaming PC, hosted APIs), and every app needs a different endpoint, key, and config.

The OllaBridge answer: apps connect to one URL with one key. OllaBridge routes each request to the right compute — and can always tell you exactly where it went.

graph TB
    A[Your Apps] -->|OpenAI SDK| B[OllaBridge<br/>Control Plane]

    B -->|Auto Routes| C[Local Laptop<br/>llama3.1]
    B -->|Auto Routes| D[Free GPU Cloud<br/>deepseek-r1]
    B -->|Auto Routes| E[Remote Workstation<br/>mixtral]

    C -.->|Dials Out| B
    D -.->|Dials Out| B
    E -.->|Dials Out| B

    style B fill:#6366f1,stroke:#4f46e5,stroke-width:3px,color:#fff
    style A fill:#10b981,stroke:#059669,stroke-width:2px,color:#fff
    style C fill:#8b5cf6,stroke:#7c3aed,stroke-width:2px,color:#fff
    style D fill:#ec4899,stroke:#db2777,stroke-width:2px,color:#fff
    style E fill:#f59e0b,stroke:#d97706,stroke-width:2px,color:#fff

Key innovation: compute nodes dial out to your gateway over outbound WebSockets. No port forwarding, no VPN, no config hell — your gaming PC at home can serve your laptop at a coffee shop.

⚡ 60-Second Start

pip install ollabridge
ollabridge start

That's it — Ollama is installed if needed, a model is pulled, an API key is generated, and your OpenAI-compatible endpoint is live at http://localhost:11435/v1.

Building from source? Two commands give you the full product, backend and dashboard:

make install      # backend + frontend into .venv (never touches system Python)
make run          # gateway + dashboard at http://localhost:11435/ui

Then point any OpenAI SDK at it:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11435/v1",
    api_key="sk-ollabridge-...",   # printed at startup
)

print(client.chat.completions.create(
    model="llama3.1",
    messages=[{"role": "user", "content": "Hello!"}],
).choices[0].message.content)

Works out of the box with the OpenAI SDK, LangChain, Node.js, cURL — anything that speaks the OpenAI API. More in docs/EXAMPLES.md.

🖥️ The Dashboard

A real-time command center for your entire LLM infrastructure — models, sources, devices, and the cloud relay, all visualized live at http://localhost:11435/ui.

🎛️ Share exactly what you choose — per model

Every model gets its own independent switches: This PC · Cloud · per-app allow-lists · Routing. Nothing leaves your machine until you flip a switch, and the Published to OllaBridge Cloud panel always shows precisely what your paired apps can see — no surprises, no hidden catalog.

🔌 Every AI account in one hub

Connect OpenAI, Anthropic, Google Gemini, IBM watsonx.ai, Azure OpenAI, AWS Bedrock, Groq, Mistral and more — 14 providers plus any OpenAI-compatible endpoint — with one uniform add → test → rotate → remove flow. Keys are encrypted at rest, stored on your machine, and never echoed back. New sources start local-only, private, and routing-off.

☁️ Pair once, reach your models anywhere

Cloud Relay	Device Pairing

Model Inventory	Local Runtimes

More screens: Providers Hub, Settings & classic Sources

Providers Hub	Settings

Sources (classic)

# Install frontend dependencies
make ui-install

# Development (hot-reload, proxied to OllaBridge API)
make ui-dev

# Production build (served at /ui when gateway is running)
make ui-build

Once built, the dashboard is served at http://localhost:11435/ui whenever the gateway runs. React 19 + TypeScript + Vite, TanStack Query live polling, Framer Motion animation.

🚀 Why Teams Choose OllaBridge

🧭 Local vs Cloud vs Enterprise

OllaBridge is local-first. Cloud is optional. Enterprise builds on both.

ollabridge start          # local only — no cloud required
ollabridge login          # optional: pair with OllaBridge Cloud
ollabridge sync status    # see exactly what syncs (and what never does)
ollabridge sync disable   # turn all cloud sync off
ollabridge logout         # unpair and delete credentials

The privacy contract (docs/PRIVACY.md, docs/CLOUD_SYNC.md): prompt content, conversation history, provider API keys, RAG documents, and persona memory are never synced unless you explicitly enable each one — and they are all off by default.

🩺 Trust, Verified — ollabridge doctor

You should never have to hope your AI infrastructure works. Verify it:

ollabridge doctor             # all checks, human-readable or --json
ollabridge doctor relay       # WSS connect, register, ping/pong, reconnect
ollabridge doctor e2e         # full request path with latency breakdown
ollabridge doctor security    # secrets at rest, permissions, auth, CORS

Relay:
  ✅ WSS connection established — wss://app.ollabridge.com/relay/connect
  ✅ Device registered — hello accepted
  ✅ Model list sent — 1 models
  ✅ Ping/pong — app-level heartbeat OK
  ✅ Reconnect test — second connection accepted

Every request carries a request_id (X-Request-ID), and the trace store records routing metadata only — never prompt content:

ollabridge traces list        # which model answered, was cloud used, latency
ollabridge route explain coding   # which backend WOULD serve this alias, and why

Policy aliases (local-private, fast, cheap, best, coding, …) are explicit and explainable — local-private is fail-closed: local devices only, no external providers, no relay, no prompt logging. Details: docs/RELAY_VERIFICATION.md.

🔑 BYOK Providers

Secure BYOK routing across your authorized devices and workspaces — keys encrypted at rest (OLLA_SECRET), never printed, never synced by default:

ollabridge providers add anthropic     # choose: local-only / cloud vault / org vault
ollabridge providers test anthropic    # validates the key — no tokens spent
ollabridge providers rotate anthropic  # replace key, stamp rotation time
ollabridge providers list              # keys always redacted

Supported: OpenAI, Anthropic, Gemini, Azure OpenAI, AWS Bedrock, Groq, OpenRouter, Hugging Face, DeepSeek, Mistral, Together, Fireworks, plus any generic OpenAI-compatible endpoint. Free-tier alias routing (free-best, free-fast) via the Providers Hub.

🏗️ Architecture

• Gateway (control plane) — FastAPI server exposing /v1/*, the dashboard, and admin APIs
• Nodes (compute plane) — Ollama/HomePilot/provider adapters that dial out over WebSockets
• Cloud relay (optional) — outbound-only WSS bridge so your devices are reachable from anywhere, tenant-isolated per account

Deep dives: docs/ARCHITECTURE.md • docs/SCALING.md • docs/THREAT_MODEL.md

🤖 AI-Agent Native

OllaBridge ships an MCP server so AI agents can manage your infrastructure — create enrollment tokens, list runtimes, check health — through the Model Context Protocol:

ollabridge-mcp        # stdio MCP server for Claude Desktop, IDEs, agents

See docs/MCP.md.

📚 Documentation

🗺️ Roadmap

• Control plane + dial-out node architecture (no port forwarding)
• Web dashboard (Broadcast Tower visualization)
• Optional cloud pairing + relay with end-to-end verification (doctor relay, doctor e2e)
• Explicit, metadata-only cloud sync with privacy-first defaults
• BYOK provider hub with encrypted key storage, rotation, and redaction
• Policy routing with explainable aliases (route explain)
• Request tracing without prompt content
• Multi-provider free-tier routing (score-based selection)
• HomePilot persona integration • MCP server for AI agents
• 🚧 Cloud encrypted vault & organization vault APIs
• 🚧 Relay streaming (delta/done frames end to end)
• 🚧 Prometheus metrics • more runtimes (vLLM, llama.cpp, LM Studio)

Full phased plan with shipped/planned status: docs/ROADMAP_ENTERPRISE.md

🤝 Contributing

We welcome contributions! Areas we'd love help with: runtime adapters (vLLM, llama.cpp), monitoring/metrics, security reviews, docs.

1. Fork the repo
2. Create a branch (git checkout -b feature/amazing)
3. Make your changes and add tests (make install-dev && make test)
4. Submit a PR

📄 License

Apache License 2.0 — see LICENSE

🙏 Built With

FastAPI • Ollama • WebSockets • SQLModel • React + Vite

💬 Support

📖 Documentation • 🐛 Report a Bug • 💡 Request a Feature • 💬 Discussions

🌟 Star History

If OllaBridge helped you, give it a star! ⭐

Compatible with HomePilot

OllaBridge is the recommended gateway for HomePilot personas. Route persona:* and personality:* models to HomePilot while serving local LLMs via Ollama — all through a single OpenAI-compatible endpoint. Setup: docs/HOMEPILOT.md.

It also relays the avatar session: 3D Avatar Chatbot opens one socket to OllaBridge and reaches HomePilot through it — locally as a straight proxy, and through Cloud over the relay link to your own machine. The browser presents OllaBridge's credential and never holds HomePilot's, and there is nothing to configure beyond enabling HomePilot. See docs/AVATAR_BRIDGE.md for the whole workflow.

Connect 3D Avatar Chatbot for an immersive VR persona experience with lip sync, gestures, and voice.