Frequently asked questions

No. Grist executes in your local working tree — the prompt goes out to the model, the source does not. Tools run locally, diffs are produced locally, and Grist does not train on your code. The gateway handles accounts, keys, and routing; it never receives your files.

The harness itself is free for personal and noncommercial use; commercial use needs a license. Inference runs on your own provider key, billed by your provider at provider prices, with zero markup from Grist. Every key carries a hard spend cap: a request that would exceed it returns a 402 instead of a bill. Teams and enterprise deployments are arranged directly — see enterprise.

The client never names vendor models. It names a rung — cheapest, medium, frontier, premium — and the gate picks the cheapest rung that can handle the task. Each rung resolves to a model at the gateway, and you can override what a rung points to in your dashboard without changing any client.

A plain CLI bills every task like it's hard, starts every session from zero, and leaves spend to chance. Grist prices per task (the gate routes each run to the cheapest capable rung), remembers what was verified across sessions, navigates by a code map, delegates to named specialists, and enforces hard spend caps — the harness around the agent loop is the product.

Memory stores only proven outcomes: tests passed, work you approved, corrections you made, checks that failed. Claims an agent merely made are never stored. Yes, you can delete it: the desktop app has a memory browser with a forget button per record, and by default memory is a plain JSON file in the project (.grist/memory.json) that you can read, diff, or delete yourself. See verified-outcome memory.

Yes. The enterprise deployment runs the harness inside your VPC on your own provider key, with shared team memory, a team code map, and per-key usage attribution. The fastest start is a measured concierge pilot with one team. Details on the enterprise page.

The code is public at github.com/grist-ai/grist under the PolyForm Noncommercial license — free for personal and other noncommercial use; commercial use needs a license from Grist. The agent loop builds on OpenCode, an MIT-licensed open-source core, and those portions remain MIT.

The CLI installs via npm (npm install -g grist-ai) and runs on macOS and Linux. The desktop app is macOS today, and it auto-updates once installed. Run grist doctor any time to check your setup.