archestra

Why we included this project

Archestra is a self-hostable platform that gives an organization one governed entry point for AI work, whether that means an internal chat for non-technical staff, a proxy so Claude Code, Codex, or Cursor can reach models with a single token, or a registry that hands agent builders controlled access to MCP tools. A single deployment routes requests to any provider, from Anthropic and OpenAI to Azure, Bedrock, or DeepSeek, under virtual API keys and per-team cost limits. The governance layer is where it stands apart from a plain gateway: SSO and role-based access, sandboxed code execution, and layered guardrails ship with OpenTelemetry traces and Prometheus metrics rather than being bolted on later. For platform and IT teams that need employees to get practical, observable access to AI tools without giving up control, that combination is worth a close look. One caveat: the code is source-available under AGPL-3.0, with a separate enterprise tier for the restricted components.