Claude Code writes the agent. Trinity keeps it from wandering off.

What it does

Trinity is an agent orchestration platform that runs each AI agent inside its own isolated Docker container. It provides a web console for fleet management, real-time observability, cron scheduling, agent-to-agent delegation, and tamper-evident audit trails. You bring the agents — it handles the runtime, memory, and governance.

The interesting bit

The project treats “sovereign” infrastructure seriously: it is Apache 2.0, independently pentested to Grade A, and designed so that even an AI agent can install itself using a deterministic runbook. The v0.9.0 release adds bidirectional A2A, a client-facing Workspace surface, and automatic database backups, suggesting the maintainers actually run this in anger.

Key highlights

• Each agent is containerized and isolated; the platform tracks health, costs, and execution history fleet-wide.
• Native integration with Claude Code via MCP and a plugin marketplace that scaffolds, deploys, and iterates agents.
• Tamper-evident audit trails and scheduling built in, not bolted on.
• SQLite for local evaluation, PostgreSQL for production, with a migration path via the Trinity Ops Agent.
• Independently pentested (UnderDefense Grade A) and Apache 2.0 licensed, including commercial use.

Caveats

• The README still contains TODO comments for missing video assets, and private git submodules that are skipped automatically — signs of a project still tidying up its public face.
• Production deployments require PostgreSQL; SQLite is explicitly labeled as dev-only, so plan a database migration before going live.

Verdict

Teams already using Claude Code and wanting to turn prototypes into scheduled, audited production workloads should look here. If you just need a single chatbot wrapper or don’t want to manage Docker, this is overbuilt.