plan mode
Read-only investigation before any edit. Press Shift+Tab to enter it.
Zone runs on your machine with your own API key. No subscription, no rate limits, and the exact dollar cost of every run.
npm install -g zone-ai-agent ~/catalora zone ▸ plan mode · investigate read 14 files · 3 searches · tsc --noEmit ▸ applied src/repricing/rules.ts +48 −6 src/repricing/rules.test.ts +112 passed tsc --noEmit · vitest 41/41 ────────────────────────────────── in 62,410 out 3,880 cached 51,220 sonnet-4.6 1m 48s $0.11
Subscription agents put your usage behind a flat price and a rate limit. You never see what a task actually cost, and the meter stops when they say so. Zone talks to the provider API with your key, so you pay per token and every run reports its own bill.
| Zone | Subscription agents | |
|---|---|---|
| billing | Per token, on your own key | Flat monthly price, one vendor |
| limits | None — never blocked mid-session | Rate-limited on rolling windows |
| cost | Exact dollars per run, logged locally | Hidden behind the plan |
| where it runs | Your machine | Their cloud |
| source | AGPL-3.0, readable and forkable | Closed |
To be exact about what stays put: your prompts and the files you reference still go to whichever model provider you chose — that is what a coding agent does. What Zone removes is everything in between. No account with us, no proxy server, no telemetry, no third party holding your key.
Real Zone runs: a small edit is about $0.02, an Astro site scaffold $0.09, a typical cached feature or refactor $0.20–0.50, a large multi-file refactor around $0.90. Move the sliders to your own habits.
Anthropic rates as of June 2026: Sonnet 4.6 at $3 / $15 per MTok, Opus 4.8 at $5 / $25, cache-hit input at $0.30 — a 90% discount. Prompt caching is on by default, so a long multi-turn run gets cheaper per turn, not more expensive. Run /cost or /metrics for your own history; full logs sit in ~/.zone/cost-logs/.
In plan mode Zone investigates read-only first — it opens files, searches, and runs your build, test, and typecheck commands — then shows a step-by-step plan and waits. Approve it, refine it, or reject it before anything is written.
Edits are validated against the syntax tree, staged in memory, and typechecked on the staged tree. They flush all or nothing, and roll back on failure. Your working tree never sits in a half-broken state.
Every run ends with its exact tokens, elapsed time, model, and dollars. Set a daily cap with /limits and Zone stops before it crosses the line.
Read-only investigation before any edit. Press Shift+Tab to enter it.
Staged, typechecked, flushed all-or-nothing, auto-rollback on failure.
Resume any run where you left it with --resume or /sessions.
Anthropic and OpenAI, the right model per task. Switch with /model.
/undo restores files to their pre-run state, and warns if you edited them since.
.zone/hooks.json defines veto rules before a tool runs and side-effects after.
Point .zone/mcp.json at stdio servers; their tools register automatically.
Inline a file with @path, attach a screenshot with /image.
npm install -g zone-ai-agent
export ANTHROPIC_API_KEY=sk-ant-... # or OPENAI_API_KEY
cd ~/your-project && zone You can skip the environment variable entirely and add keys from inside the TUI with /keys. No .env file is needed for personal use.
git clone https://github.com/BedreddinErkan/zone && cd zone npm install && npm run build && npm link
Node.js 22+
npm 10+
Windows: run it inside WSL 2