Airlock is an enterprise LLM proxy built on LiteLLM. It gives AI coding tools a single OpenAI-compatible endpoint while adding guardrails, routing, observability, and operational controls.
- New to Airlock? Start with Installation, then Configuration, then Connecting AI Tools.
- Operating a deployment? See Operations, Troubleshooting, and the Admin API.
- Integrating a client? Consult the live API reference, response-header reference, and routing guide.
- Evaluating the design? Read the architecture overview and system diagram.
| Capability | Start here |
|---|---|
| Provider routing and model aliases | Routing |
| PII redaction, safe tool hydration/egress, keyword, semantic, and response controls | Guardrails |
| Provider protection and admission | Rate limiting |
| Logs, metrics, tracing, and response transparency | Observability |
| Batch workloads | Batch processing |
| MCP tools | MCP servers |
| Optional FathomDB search/analysis and per-client erasure | Fathom storage and Operations |
| Opt-in Admin recovery actions and same-host read-only fleet view | Admin API |