Cerebro Engine — flow diagram
A multi-agent engine (LangGraph) that turns a task description into a reviewed, evaluated deliverable, autonomously: it clarifies ambiguities, plans, designs, implements, reviews — and when hooked to a real project, integrates the code and verifies the result.
At a glance
- Type
- Multi-agent engine, built on LangGraph
- Interface
- Command line only
- Actors
- Six agents in sequence, plus integrator and verifier when the engine is hooked to a real project
- Models
- One per agent, configurable independently — local or remote
- Human in the loop
- Yes: questions and answers over the CLI, and most stages can be restarted on demand
- Status
- Working
How it works
- Clarifier — takes the task description and asks the human whatever is ambiguous, before anything else starts
- planner — turns the clarified task into a plan
- organizer — designs how that plan will actually be carried out
- executor — implements it
- reviewer — reviews the result
- finalizer — closes the run and hands back the finished deliverable
- integrator and verifier — only when the engine is hooked to a real project: the code is integrated and the result verified
Decisions and constraints
- Provider-agnostic, one agent at a time — each agent runs its own configurable model, local or remote. The goal is what matters, not the model
- The human can restart most stages — planner, organizer, executor and reviewer can be re-run on demand; Clarifier and finalizer can't
- The conversation happens only over the CLI — the engine asks and the human answers there and nowhere else
- Integration is optional — integrator and verifier come into play only when the engine is pointed at a real project; on its own it stops at the reviewed deliverable