astaxie/TokenHub

TokenHub – Enterprise‑grade AI model gateway & governance

What it does
TokenHub sits in front of any OpenAI‑compatible LLM provider (OpenAI, Azure, Anthropic, Gemini, DeepSeek, Qwen, local models, etc.) and adds a thin‑but‑powerful governance layer for enterprises.
It lets you:

• Route calls to the best model for a scenario, cost, health or fallback policy.
• Control who can use which tokens – project‑scoped API keys, team‑level permissions, quotas and concurrency limits.
• Save money – automatic caching, model‑selection rules and budget caps.
• Reconcile bills – compare internal usage logs with provider invoices so finance can explain spend.

Who it’s for

• Users – developers who just need an API key to call a model and see their own usage.
• Team leaders – manage projects, members, cost attribution and reports.
• Administrators – configure providers, routing policies, RBAC, audit trails and cost controls.

Key technical bits

• Written in Go 1.26 with a SQLite‑first deployment option; PostgreSQL can be used for multi‑instance scaling.
• Provides OpenAI‑compatible endpoints (/v1/chat/completions, /v1/embeddings, /v1/images/generations, etc.) plus Anthropic‑style messages and token‑count APIs.
• Supports Docker‑Compose or native systemd installation; release bundles are version‑checked and can be rolled back from the admin console.
• Includes 150+ provider templates and adapters for the major commercial LLM services; custom upstreams can be added via the generic OpenAI‑compatible interface.
• UI offers light/dark mode, role‑aware navigation and analytics dashboards (usage per user, project, model, cost centre).

Getting started

1. Native install – run a curl‑downloaded script that installs a systemd service and prints an admin password.
2. Docker‑Compose – copy .env.example, fill in secrets, run ./deploy/install.sh.
3. Open the console at http://localhost:3000 (admin login admin).

Documentation

• Architecture overview, user/leader/admin guides, deployment instructions and contribution guide are all in the docs/ folder.
• The project is Apache‑2.0 licensed and welcomes contributions.

---

TokenHub is a genuine, production‑oriented open‑source project that tackles the real‑world problem of governing AI model consumption in large organisations.