generalstaff.io · pre-launch
One desk for every AI subscription you already pay for.
GeneralStaff is a command desk for people who run two or more AI coding tools. You talk to one orchestrator conversation. It hands bounded work to whichever of your signed-in subscriptions fits: Claude Code, Codex, Cursor and similar. You accept a change only after it has passed tests and a different model has reviewed it.
Join early access See the first ten minutes
Five seats at the desk: Orchestrate · Build · Review · Verify · Fast assist
You
Move session storage from the in-memory map to Redis, behind the existing interface.
Orchestrator
Bounded to src/session/. Assigned to the Build seat. Review goes to a different model.
Evidence · one bounded change
src/session/store.ts
- const session = memory.get(req.cookies.sid) + const session = await sessions.get(req.cookies.sid)
- ✓ Tests passing
- ✓ Reviewed: approved, second model
- Audit recorded
- Rollback available
01 · The problem
You pay for the models.
You still do the managing.
The juggle
Two or more AI coding subscriptions, each with its own window and its own idea of your project. You are the integration layer between tools you already pay for.
The guesswork
You pick a subscription by habit, and the context you pasted into the last window is stale by the time the answer comes back.
The trust gap
Output arrives looking finished. Unless you test it and read every line yourself, you accept changes on faith, and the model grades its own homework.
02 · How it works
The first ten minutes.
-
minute 0
Install the desk
A desktop app. No terminal, no config files.
-
minute 1
It finds your tools
GeneralStaff detects the AI coding subscriptions signed in on your machine.
-
minute 3
Pick a repo
One desk for every project. The orchestrator reads the repository you point it at.
-
minute 5
Give one task
Describe the change once, in the orchestrator conversation. It bounds the work and assigns it to the seat that fits.
-
minute 10
Read the evidence
The diff, the test result, the reviewer’s verdict. Accept it, send it back, or roll it back.
03 · The verification gate
Nothing merges on a model’s say-so.
Tests run before acceptance, and a different model than the one that wrote the change reviews it. The builder never grades its own work. Each accepted change arrives with its evidence attached, plus a way back.
-
The diff
What changed, bounded to the task you gave.
-
The test result
Run before you accept the change.
-
The reviewer’s verdict
A second model reads the change and signs it, or sends it back.
-
The audit record
Each decision logged, with rollback when you want it.
04 · Your subscriptions, your machine
The desk keeps the books.
The subscriptions stay yours.
GeneralStaff runs locally, with the logins on your machine. It never holds your credentials and never resells AI usage. The subscriptions you pay for do the work.
- Runs locally on your machine
- Uses your own signed-in logins
- Never holds your credentials
- Never resells AI usage
Works with your signed-in tools
- Claude Code
- Codex
- Cursor
- …and similar subscriptions
05 · Open-source core
The engine is open.
The experience is the product.
The engine
Open source under the AGPL. Read all of it: the orchestration, the gate, the audit trail.
The desktop app
The paid product is the experience: no terminal, no config files, one desk for every project.
06 · Early access
Join the list.
GeneralStaff is pre-launch. Leave an address and you will hear when seats open. No prices today. The list hears first.
A mock form for now. Nothing leaves this page.