Otari is an open-source LLM gateway and control plane you self-host. It puts one OpenAI-compatible endpoint in front of 40+ providers, so your applications keep a single stable API while you swap models behind it. It also speaks the Anthropic Messages API, so Claude-native clients work without changes.

Because Otari runs on your own infrastructure, prompts and responses never reach Otari's servers. Only metadata is stored. That suits teams with data-residency, privacy, or compliance requirements that rule out a hosted third-party hop.

What it does

• Routes across 40+ providers through one OpenAI-compatible (and Anthropic-compatible) endpoint
• Issues virtual API keys with per-key budgets and spend caps
• Tracks usage and attributes cost per key
• Applies request-level guardrails, including prompt-injection detection
• Ships built-in server-side tools (sandboxed code execution, web search) so open-weight models reach parity with managed frontier tooling

Otari is built on any-llm, which calls each provider through its official SDK instead of reimplementing the interface. That keeps compatibility close to the source and reduces breakage when providers change. It is released under Apache 2.0.

For teams already running a proxy, migration is direct: point your existing OpenAI-compatible SDK at Otari's URL, and your keys and environment carry over.

Self-host it from the repo, or use the hosted platform at otari.ai if you would rather not operate it yourself.