Otari

About

OpenAI-compatible LLM gateway you run yourself, by Mozilla.ai

Otari sits between your apps or agents and the model providers. It authenticates each request, resolves the provider credential, enforces the budget before dispatch and logs usage afterwards, with routing policies for failover and weighting between models. Provider calls go through Mozilla.ai's any-llm library. It runs standalone with Docker and your own database and keys, or you can connect a data plane to the hosted otari.ai platform. For coding agents there is a separate otari CLI (brew install mozilla-ai/tap/otari) with hooks and a Claude Code import, plus optional guardrails, web search, MCP and code execution. The dashboard shows spend by model, keys, members, budgets and a playground for comparing models. Repo: https://github.com/mozilla-ai/otari.

Features

• One OpenAI and Anthropic compatible endpoint for 40+ providers
• Revocable API keys scoped by user, workspace and model
• Budget checks before spend, usage records after
• Routing policies for failover and weighting
• Dashboard with spend, usage and a model playground
• Coding agent hooks and Claude Code import
• Runs with Docker or connects to hosted otari.ai

Install