Work / BigTerminal
BigTerminal
The execution engine inside BigBrain: a queue of software goals, each run by a supervisor model and isolated worker agents across cloud and local providers. Nothing counts as done until its checks pass.
What it does
You give it a goal from the browser or the terminal, or queue several. Each run gets a supervisor model, workers in their own Git worktrees and a reviewer that is a different model from the writer. Routing chooses across Claude, OpenRouter and local Ollama, and an operator pin can overrule it. A rate-limited provider parks the run until its reset time instead of failing it. Every attempt records its model, tokens, time and cost, and nothing completes until the Definition of Done passes.
- Give it a goal from the browser or the terminal
- Routing picks a provider and model; an operator pin can overrule it
- A supervisor model plans the work as structured directives
- Workers carry it out in isolated Git worktrees
- A different model reviews the change
- Definition-of-Done checks decide whether it counts
- The result lands with its model, tokens, time and cost recorded
How it works
BigBrain is a Node and Express control plane with a React interface. Runs, attempts, events and budgets are kept durably in SQLite, and the live run streams to the browser over Server-Sent Events, resuming from a cursor after a reconnect. A supervisor model issues structured JSON directives. Workers carry them out in isolated Git worktrees, with safety enforced at the Git level rather than in the prompt. Independent product and regression proofs decide whether work is accepted, because an earlier task once replaced a richer feature with a worse one while every implementer test stayed green.
Skills shown
- Databases & SQL
- Runs, attempts, events and budgets are kept durably in SQLite.
- DevOps & CI/CD
- Each worker gets its own Git worktree, and a Definition-of-Done gate blocks completion until every check has passed.
- APIs & backend
- A Node and Express control plane that issues structured JSON directives and parks runs on provider rate limits.
- Real-time & streaming
- Live runs stream to the browser over Server-Sent Events and resume from a cursor after a reconnect.
- AI, agents & ML
- Routes work across Claude, OpenRouter and local Ollama models, with a supervisor, isolated workers and a reviewer that is a different model.
- Testing & QA
- Independent product and regression proofs decide acceptance; the implementer’s own green tests are not enough.
Worth knowing
- Routes each task across cloud and local model providers, and records why a route was chosen or overruled.
- Runs worker agents in separate Git worktrees, so parallel AI writers cannot overwrite each other or the operator's work.
- Refuses completion until every Definition-of-Done item has passed with evidence. Green implementer tests alone are not accepted.
- Parks a run when its provider is rate-limited or out of capacity, until that provider's reset time, instead of failing it or burning retries.
- Turn gating cut a run that had spent 21 supervisor turns, mostly waiting, down to 2.