Trinity
Trinity is a self-hosted agent orchestration platform with Docker isolation, fleet observability, scheduling and RBAC for autonomous agents.
Pros
- + Per-agent Docker isolation with configurable resource limits
- + Fleet observability: graph topology, live status, cost and health
- + Cron scheduling coordinated via Redis distributed locks
- + Four-tier RBAC, audit logging and encrypted credential storage
- + Agent-to-agent delegation and MCP integration
Cons
- − Polyform Noncommercial license — commercial use needs a paid license
- − Multi-service Docker Compose stack is heavier to operate
- − Tied to Claude Code or Gemini CLI as agent runtimes
- − SQLite backend may constrain very large fleets
What Is Trinity
Trinity is an open-source platform for deploying, managing and monitoring fleets of autonomous AI agents on your own infrastructure. It gives organizations governance, observability and scheduling without requiring data to leave their security perimeter.
It exists because the existing options for production agents are all flawed: SaaS platforms compromise data sovereignty, building in-house takes six to twelve months of engineering, and most frameworks have no fleet-management layer at all. Trinity’s answer is real-time observability, fleet-wide scheduling, agent-to-agent delegation and complete audit trails — all running on infrastructure you control.
Architecture
Trinity is a Python/FastAPI backend with a Vue.js 3 front end, deployed as a Docker Compose stack. Each agent runs in its own isolated Docker container with dedicated resources.
| Service | Role |
|---|---|
| FastAPI backend (8000) | API and control plane |
| Vue.js frontend (80) | Dashboard and admin UI |
| MCP server (8080) | Tool integration layer |
| Redis | Secrets and distributed locks |
| SQLite | Relational data store |
| Vector | Log aggregation |
| OpenTelemetry collector | Optional metrics export |
Agent isolation is container-level: each agent gets its own Docker container with configurable resource limits, an optional read-only mode to protect source code, and an elevated full-capabilities mode for agents that need system-level access. Scheduling is cron-based through a dedicated scheduler service that uses Redis distributed locks so multiple scheduler instances coordinate safely. Persistence spans SQLite for relational data, file-based persistent memory, encrypted git storage for credentials, and Redis for state coordination. Agent runtimes are Claude Code (Anthropic) or Gemini CLI (Google).
Key Features
Fleet observability
Graph topology view with live agent status, success rates, cost tracking, timeline views, host telemetry, health monitoring and OpenTelemetry metrics export.
Agent runtime
Docker isolation, multi-runtime support, per-task model selection, custom dashboards, playbooks, persistent memory, runaway prevention via max_turns, guardrails and execution retry.
Orchestration
Agent-to-agent communication, parallel task execution, shared folders, cron scheduling, MCP integration, WebSocket event streaming and channel adapters for Slack and Telegram.
Operations
Template-based deployment, encrypted credentials, per-agent GitHub tokens, four-tier RBAC, audit logging, ephemeral SSH access, public agent links and a mobile admin PWA.
What’s Missing
The biggest caveat is licensing: Trinity uses the Polyform Noncommercial License 1.0.0. It is free for personal, research, educational, non-profit and hobby use, but commercial use requires a separate paid license — so businesses cannot simply adopt it for free. Operationally it is the heaviest tool in this category: a multi-service Docker Compose stack (backend, frontend, MCP server, Redis, Vector, optional OpenTelemetry) is more to deploy, monitor and keep healthy than a single process. Agent runtimes are limited to Claude Code or Gemini CLI, so other providers are not first-class. The SQLite relational store is fine for typical fleets but may become a constraint at large scale or under heavy concurrent write load. Hobbyists wanting something lightweight, or commercial teams unwilling to license, should look elsewhere.
Who It’s For
Trinity is for engineering teams and the managers running them who need to operate a fleet of autonomous agents on sovereign infrastructure — with observability, scheduling, RBAC and audit trails — and who either qualify for noncommercial use or are willing to buy a commercial license. It is a serious platform for serious fleet operations, not a quick personal assistant.