Build AI Systems on governed rails

Your agents, assistants and integrations call LLMs, MCP tools, A2A agents and skills through Brutor — and every AI System you ship gets guardrails, budgets, audit and runtime assurance without changing your code. Looks friendly. Works like a beast.

Quickstart — run Brutor Your first governed LLM call

Pick your path

I'm building an agent or app

Point your existing OpenAI or Anthropic SDK at Brutor with a one-line base-URL change, call MCP tools through a governed endpoint, or build a full custom portal on the Portal API.

Overview for developers →

I'm running AI Systems in production

Declare each agent or assistant as an AI System, and Brutor watches it for you: liveness, behavioural drift, contract conformance and a health verdict that never reports green on absent evidence.

How assurance works →

I'm installing and operating Brutor

Run the full platform with Docker Compose in about ten minutes, then take the governance and observability tours to see what the control plane does for you.

Installation guides →

Your AI System, end to end

What is an AI System?

The unit you govern and assure: the agent, assistant or integration as a whole — the models, MCP tools, skills and agent identities it uses, bound together, with every run attributed back to it.

Meet the AI System →

Develop → Deploy → Govern → Observe → Assure

One journey takes a system from first prototype to production evidence: build it on governed rails, promote it through the lifecycle gate, and let the run ledger prove how it behaves.

Follow the journey →

What you build with Brutor

Agents

LangGraph, Claude Agent SDK or plain HTTP — agents call LLMs and MCP tools through the gateway and inherit guardrails, budgets and audit automatically. Build an agent →

Custom portals & chat clients

The Portal API gives you end-user auth, workspaces, model discovery, streaming chat, tool approvals and notifications — bring your own UI. Build a portal →

MCP servers & tools

Register any MCP server once and every governed client — Claude Desktop, Goose, your agents — gets it with access control and tool-call policies. Register a server →

Setup & provisioning scripts

Everything the Admin UI does is a REST API. Load users, groups, models, guardrails and budgets from an idempotent script. Write a setup script →

The platform

Governance

Resource groups, guardrails, argument & semantic policies, budgets, quotas, approvals and compliance tagging.

AI System Assurance

The run ledger, liveness, behavioural drift, contracts, replay and the health verdict — evidence that each system still does what you approved.

Observability

Mission Control dashboards, proxy & usage logs, audit trail, Prometheus metrics and Grafana.

API Reference

Runtime & Portal APIs on the Core Proxy (:8100), Admin API on the Control Plane (:5050).

Architecture

How the Rust core proxy, Python control plane, Admin UI and User Portal fit together.

Compliance

The Asset Register, EU AI Act obligations by system, role and date, and the ISO 42001 management-system layer.