Self-host Otari

Otari is an LLM gateway from Mozilla.ai that sits between applications and model providers. It routes requests to 40+ providers through a unified OpenAI-compatible endpoint, manages provider credentials securely behind the gateway, issues revocable virtual API keys with user/workspace/model scope, enforces budgets before requests are dispatched, and records usage after settlement. It supports local routing policies for failover, weighted selection, and learned routing. Optional features include sandboxed code execution, web search, MCP integration, inference guardrails like prompt-injection detection, and file understanding for local models. The gateway is deployed via Docker or Docker Compose with PostgreSQL for production, and includes a web dashboard for managing providers, models, routing, keys, members, budgets, and observability. The CLI tool (installable via Homebrew) adds agent guardrails for coding agents like Claude Code without requiring the server.