archestra

Archestra is an all-in-one open-source enterprise AI platform designed to centralize AI access and management. It provides a chat interface for non-technical users with projects, MCP apps, and Slack/MS Teams/email front-ends. For developers, it offers an LLM and MCP portal with a single token for Claude Code, Codex, and Cursor, plus an LLM gateway supporting many providers (Anthropic, OpenAI, Azure, Bedrock, DeepSeek) with cost limits, virtual API keys, and dynamic model routing. The MCP gateway uses OAuth and On-Behalf-Of for user-context tool execution, and an A2A gateway handles agent-to-agent triggers. A private MCP registry lets teams publish their own tools, and an MCP orchestrator with a Kubernetes operator manages environments and self-serve promotion. The agent runtime supports scheduled/email/webhook triggers, sub-agent delegation, reusable skills, sandboxed code execution, and a K8s-native filesystem. RAG knowledge bases connect via connectors. A mini app builder enables custom apps. Security features include deterministic guardrails for tool calls, Dual-LLM verification, Lethal Trifecta protections, SSO (OIDC, SAML, Okta, Entra), RBAC with role mapping and team sync, and secrets management. Environments support per-env egress policies and cost limits. Observability is built-in with OpenTelemetry traces, Prometheus metrics, logs, and per-team cost tracking. Deployment is via Docker or Helm/Kubernetes. The project is open-core (AGPL 3.0 / Enterprise), free for teams under 30 users, and has stable and beta release channels. It is a Linux Foundation and CNCF member.