Open-source infrastructure for enterprise AI

Move AI into productionwithout losing control of data,access, and actions.

Built for platform teams rolling out AI across the company.

Runs in your Kubernetes clusterModel and framework agnostic

2. Platform architecture

One production layer between AI and your systems.

AI becomes useful when it can tap into your data, call your tools, and take action—not just generate text. Archestra gives every agent and AI client one controlled gateway to your systems.

Explore the Archestra platform

AI clients and agents

Claude · Codex · Cursor · n8n · Internal agents

Archestra

LLM proxy · MCP gateway · Identity · Policy · Execution · Audit · Cost Control

Your infrastructure

Models · MCP servers · Knowledge · Business systems

— Santiago Barclay, Head of Cybersecurity at Akua

3. Production controls

Control what AI can access and do.

Safe AI is not AI that never makes a mistake, but the one whose access and actions are bounded, enforced, and visible.

Every action has an identity

Connect Entra ID, Okta, OIDC, or SAML. Archestra records who initiated each tool call and resolves user credentials at runtime.

Explore identity and access

Policies are enforced before execution

Control tool access, approvals, isolation, and network egress. Block unauthorized actions and unsafe data flows before they reach your systems.

Explore AI security and guardrails

Every request is traceable

Trace model requests, tool calls, policy decisions, and cost. Export telemetry through OpenTelemetry and Prometheus.

Explore observability

4. Quickstart

Deploy your way

Run locally, develop from source, or deploy to Kubernetes.

Explore deployment options

#Requires: Docker

$docker pull archestra/platform:latest

docker run -p 9000:9000 -p 3000:3000 \ -e ARCHESTRA_QUICKSTART=true \ -e ARCHESTRA_BETA=true \ -v /var/run/docker.sock:/var/run/docker.sock \ -v archestra-postgres-data:/var/lib/postgresql/data \ -v archestra-app-data:/app/data \ archestra/platform

#Then open http://localhost:3000

#Full guide: Deployment Guide

5. Works with your stack

Keep the tools your teams already use.

Engineers keep Claude Code, Codex, Cursor, Copilot CLI, n8n, and their preferred frameworks. Other teams can work directly in Archestra using Chat, Apps, RAG, and shared Projects.

Archestra gives your AI tools a shared control layer, integrated with your existing models, identity providers, and observability stack.

View tools

Models:

OpenAI · Anthropic · Azure OpenAI · Vertex AI · Bedrock · Self-hosted

Infrastructure:

Kubernetes · Helm · Postgres

Observability:

OpenTelemetry · Prometheus · Grafana · Splunk

6. Entry points

Start with what's blocking production.

Enterprise MCP infrastructure

Give MCP servers one registry, gateway, identity layer, and audit trail.

Explore MCP

LLM proxy

Give teams one endpoint for approved models. Route requests across providers, keep API keys server-side, set spend limits, and trace every request.

Explore the LLM proxy

AI agent runtime

Move agents off laptops and run them as managed, sandboxed workloads.

Explore Agents

Archestra is a modular platform.Use it all or just the parts you need - we won’t be offended.

Move your first use case into production.

Bring us the agent, MCP deployment, or access problem that is blocking your rollout.

Book a technical demo

FAQ

What does Archestra cost?

Archestra is licensed under AGPL-3.0 and free to self-host. Chat, agents, the MCP gateway, LLM proxy, orchestrator, and observability are included. SSO and OIDC, RBAC, white-labeling, and advanced knowledge bases with access control require an enterprise license. Teams under 30 users get that license free.

Is Archestra open source?

Yes - AGPL-3.0, the whole platform, not an SDK. The repo is on GitHub, and the quickstart runs in about three minutes.

We’re a small team. Do we need to talk to sales?

No. Under 30 users, everything is free, including the enterprise features. Deploy and go. If you grow past that, email us - you’ll know when.

How is Archestra different from LiteLLM?

LiteLLM is an LLM proxy - very good at routing model traffic. Archestra includes its own LLM proxy and covers the layers around it: MCP infrastructure, agents, chat for non-technical teams, guardrails. Already run LiteLLM? It plugs in too; Archestra is a modular platform.

How is Archestra different from an MCP gateway?

A gateway routes MCP traffic. Archestra includes one, plus the registry of 900+ evaluated MCP servers, agent runtime, chat, knowledge bases, and cost tracking that turn it into a platform your whole company can use.

Couldn’t we build this ourselves?

You could - and several of our customers started that way. The gateway and proxy are the easy part; the policy engine, credential isolation, and auth flows are where the real time goes.

Our engineers already live in Claude Code, OpenClaw, and Cursor. Do they have to switch?

No, they can keep their tools. Connect in one click and keep working as before. Behind the scenes, every call now goes through Archestra: one registry, one policy set, full visibility.

Can we control token spend costs?

Yes, set budgets per person, team, or agent. Virtual keys, so no shared API keys and no mystery spend. Everything lands in your Grafana. Cheaper model does the job? The proxy routes to it.