Why Trinity Lite?

Most agent frameworks start by asking you to build agents inside their runtime. Trinity Lite starts from a different reality: you already have CLI agents on your machine, and you need a local AgentOps layer for reliable handoff, recovery, independent review, and evidence-based acceptance.

The Job

Trinity Lite operates local CLI agents through a SQLite task bus:

prompt -> route -> queued task -> worker -> CLI command -> result -> review -> verification -> acceptance evidence

It is deliberately small:

• no server required;
• no model-provider abstraction in the core package;
• no credentials in project config;
• no shell string execution for agent commands;
• no need to rewrite existing CLI agents as framework objects.

Comparison

Project	Primary job	Best for	Trinity Lite difference
LangGraph	Build resilient graph-based agents	Stateful agent applications, graph workflows, production agent control flow	Trinity Lite coordinates already-installed CLI agents instead of defining agents inside a graph runtime.
CrewAI	Build role-based multi-agent crews	Autonomous role-play style teams and task delegation	Trinity Lite focuses on local CLI handoff, SQLite audit state, and explicit acceptance evidence.
AutoGen	Build programmable multi-agent conversations	Research and application frameworks for agentic workflows	Trinity Lite is a smaller operational bus for command-line tools, not a general conversation framework.
OpenHands	Run an AI software engineering agent environment	End-to-end autonomous coding in a managed workspace	Trinity Lite does not replace a coding agent; it lets multiple local agents coordinate and review each other.

When Trinity Lite Fits

Use Trinity Lite when you want to:

• connect Codex, Claude Code, Hermes, or custom CLI tools on one machine;
• keep every task, result, error, and review link in local SQLite;
• run a mock demo before configuring real agents;
• route by capability instead of hardcoded agent names;
• prove that work was reviewed and verified before acceptance;
• expose the same bus to MCP clients.

When It Does Not Fit

Do not start with Trinity Lite if you need:

• a hosted cloud agent platform;
• a full graph runtime with complex branching and streaming state;
• managed browsers, sandboxes, and remote execution;
• a provider-specific model API wrapper;
• a UI-first product.

Those can be layered around the bus later, but they are not the core package.

The Position

Trinity Lite sits between single-agent CLIs and full agent frameworks:

Codex / Claude Code / Hermes / custom CLI
        |
        v
Trinity Lite: local routing, task bus, review, verification, acceptance evidence
        |
        v
Optional layers: MCP clients, dashboards, docs sites, CI checks

That narrow scope is the product: a developer can install it, run the mock flow, then wire in one real CLI at a time without giving up local task truth. The durable differentiator is not simply launching multiple agents; it is recovering their state and accepting work only after review and verification evidence.