Findie routes every card to the model it deserves: Claude, Codex, Gemini, Cursor, Copilot, OpenCode, GLM and MiniMax on one board, with per-task budgets that hold.

Eight brains. One grove.

Claude, Codex, Gemini, Cursor, Copilot, OpenCode, GLM, MiniMax: one root system. Route every card to the model it deserves; budgets meter the sap.

● grove alive
your cursor is the light
hold anywhere · force the bloom
01/03
route

Every card picks its own branch: Opus for the hard one, Haiku for the chore, Codex when you feel like it.

Different brains,
same discipline.

Whatever model builds the card, the engine is identical: isolated worktree, serialized queue, fast-forward only. This is a real run.

A Findie agent streaming its plan and edits in real time
claude · agent sdk codex gemini cursor copilot opencode glm · minimax
The doctrine

Route per card

the model is a property of the task, not of the product. mix freely on one board.

Budgets meter the sap

per-task dollar caps, metered live per token. an agent over its cap stops on the spot.

One queue for all

whoever wrote the code, it lands the same way: rebase, revalidate, squash, ff.

Bring your own keys

your subscriptions, your api keys, sealed in the local vault. no middleman markup.

No lock-in

pass-through ids welcome: glm, minimax, openrouter. swap brains without repotting.

$ findie run --model opus-4-8 · haiku for the chores

The model is a choice.
The discipline is not.

per-card choice

Pick a brain per task. The board doesn't care; the queue treats them all the same.

metered sap

Cost caps are enforced mid-run, not reported after. Nothing spends in the dark.

swap anytime

New model tomorrow? Change one field on the card. The grove grows around it.

Download for macOS
bring your own model subscriptions
nasıl yapıldı? → rehber (tr)