antiquota

Keep your coding agent working around the clock across your own Google AI Pro accounts.

antiquota runs Google's Antigravity CLI (agy) across every Google account you have. When one account runs out of quota, the same conversation continues on the next one, with no lost context and no babysitting.

---

Why

Many people have more than one Google account with AI Pro, for example a personal and a work account. Each account has its own 5-hour and weekly quota. A long agent run burns through one in a few hours, then stops and waits.

antiquota treats the accounts you own as a single pool:

• One account is tired, the next one takes over. A spent quota walls that account until it resets. The turn continues in the same agy conversation on the next account that has headroom.
• Nothing expires unused. By default it spends the allowance that resets soonest first.
• Agents run 24/7. Give a session a goal and it keeps working through the night. When every account is spent, it sleeps until the earliest reset and carries on by itself.

With enough accounts, the agent never waits for quota.

_{One account spent its 5-hour quota and is walled until 06:02. The next account picked up the same session, and the routing log records the handoff.}

How account switching works

• Quota-aware routing. Each account's Gemini and Claude/GPT pools are tracked separately from agy's own quota panel. Readings refresh every 10 minutes and cost no tokens, even while agents work.
• Rotation you choose: soonest reset first (default), one by one, or round robin. A small reserve (3% by default) moves a session on before it fails mid-turn.
• Fill first or spread first. Stack sessions on one account until it runs out, or spread them across accounts so their quotas work in parallel. Several sessions can share one account.
• Errors are told apart. A spent quota walls an account until its reset. A rate limit while quota remains only rests it (1, 5, then 15 minutes). Temporary errors retry, then move on. A rejected sign-in parks the account until you sign in again.
• Survives network drops. If Wi-Fi goes away, sessions show waiting for network and pick the same turn back up when it returns. No account is blamed and no retry is wasted.
• Every decision is logged: which account served what, walls, rests and retries.

All accounts share one conversation store, so a handoff is a real continuation, not a restart from a summary. Only when a conversation truly cannot be resumed does antiquota start a fresh one, with its own handoff summary in front of the next message.

Run an agent all night

Goal mode keeps an agent working until the job is really done:

1. You state an objective and, optionally, a check command (go test ./..., pytest -q).
2. The agent works and claims completion only when it believes the objective is met.
3. The check command must pass, and an independent verifier (a fresh, read-only agy conversation) must agree. Otherwise the reasons go back to the agent as its next prompt.
4. Long jobs (training, big test suites) run in the background. The agent gets their log when they finish.

Combine goals with account switching and a few accounts become one agent that keeps going for days. Two sessions can also hand work to each other through a shared file, for example one that plans and audits and one that builds.

_{The compact transcript (the default) shows a goal as prompt, text, folded activity and outcome. Verbose and Full show every step.}

Everything else you'd expect

Sessions	Many sessions in parallel, per project folder; split view; search; fork, rewind, compact
Transcript	Streamed replies with Markdown and code highlighting, thinking, tool previews; compact, verbose and full modes
Safety	A permission policy decides every tool call; approvals appear in the app; plan mode only proposes
Git	A separate git worktree per session; review, commit and merge from the Changes panel
Usage	Tokens per day, per account and per model, including cached input and cache hit rate
Composer	@file mentions, attachments, dictation, !shell, # note, slash commands, prompt history
Background work	RUN_IN_BACKGROUND jobs, /loop, daily /schedule, a terminal per session, keep-awake
Desktop app	Native app with tray, notifications, dock badge and the system folder picker (Finder, Explorer, or your Linux desktop's)
CLI	antiquota -i for a terminal session that the app shows live; headless antiquota goal

Getting started

You need: the Antigravity CLI (agy) installed, and one or more Google accounts with AI Pro.

1. Install the app. Download the latest release for your platform, or build it from source (below).
   • macOS (Apple Silicon): open the .dmg and drag Antiquota to Applications. Builds are not notarized, so the first launch needs right-click → Open.
   • Linux: install the .deb or .rpm, or run the .AppImage.
   • Windows: run the .msi or -setup.exe installer. SmartScreen warns about unsigned builds.
2. Add your accounts. In Settings → Accounts choose Add account and sign in with Google. Repeat for every account. Accounts are listed by email, each with its 5-hour and weekly quota.
3. Start a session in a project folder and send a prompt, or press Goal to give it an objective.
4. Leave it running. The routing log shows each handoff.

antiquota keeps its own copy of your agy configuration (rules, skills, hooks, MCP servers). Your agy CLI and the Antigravity IDE are never changed.

Terminal

antiquota -i                        # interactive session, shared live with the app
antiquota -i --resume SESSION_ID    # continue a saved session
antiquota goal "all tests pass" --check "pytest -q" --cwd ~/code/app
antiquota login                     # add a Google account
antiquota accounts                  # quota, walls and leases per account

See docs/CLI.md for every command and slash command.

Platform support

Platform	Status
macOS (Apple Silicon)	Built, tested and used daily
Linux (x86_64, arm64)	Supported in the code and installers: tray, keep-awake through systemd-inhibit, native folder picker. Not yet tested on real machines
Windows (x86_64)	Supported in the code: tray, keep-awake, job objects for process cleanup, Explorer folder picker. Not yet tested on real machines

Reports from Windows and Linux are very welcome. Please open an issue with what worked and what did not.

How accounts stay separate

~/.antiquota/
  stack/.gemini/     antiquota's copy of your agy setup: rules, skills, hooks, MCP, plugins
  slots/<email>/     one home folder per account; its sign-in token is a file inside it
  sessions/<id>/     one session: metadata and its full event history
  state/             quota readings, walls, leases, the routing log and usage ledger

agy normally keeps its token in the OS keyring, where a second account would overwrite the first. antiquota gives every account its own home folder and has agy store its token as a file there, so accounts never collide and your own agy sign-in is left alone. All slots link the same conversation store, which is what lets a turn continue on another account.

The engine listens on 127.0.0.1 only, and every API call needs the install's token.

Build from source

Requirements: Rust 1.99+, Node 26 with pnpm 11, and just. On Linux, also the webview libraries: libwebkit2gtk-4.1-dev libappindicator3-dev librsvg2-dev patchelf. On Windows, the Microsoft C++ build tools and WebView2 (preinstalled on Windows 11). Run just from Git Bash, since the build scripts are POSIX shell.

Installers are built on the platform they are for: build a macOS .dmg on a Mac, Linux packages on Linux and Windows installers on Windows.

pnpm --dir ui install --frozen-lockfile
just build           # bundled engine + desktop app: .dmg (macOS), .deb/.rpm/.AppImage (Linux), .msi/.exe (Windows)
just dev             # engine and UI in development mode
just check           # formatting, clippy, cargo-deny, type checks, lint
just test            # Rust tests, Vitest, then Playwright end-to-end tests

Every test runs against a fake agy with synthetic accounts: no real account or quota is needed. docs/ARCHITECTURE.md describes the layers, and docs/adr/ records every design decision.

A note on accounts

antiquota only automates accounts you sign in to yourself, on your own machine. It does not create accounts, share tokens or bypass any limit on a single account: each account is used within its own quota. Use accounts you own or are allowed to use, and follow the terms of the services involved.

License

MIT