[ docs · guides ]

Enterprise deployment

Grist for teams is API-only: the harness runs inside your VPC, and every model call rides your own provider key. Grist never resells inference, never resells hardware, and never sees your source code.

What the team gets

• Shared team memory. Verified outcomes from everyone's sessions brief the whole team — a fix one engineer proves becomes context for every agent run after it.
• A team code map. One shared map of the codebase, so agents start from the team's structure and conventions instead of re-deriving them per session.
• A usage dashboard. Per-key usage and spend attribution: which team, which key, which task — the numbers a rollout is judged on.

Deployment shape

The gateway components run in your cloud. Execution stays where it always is in Grist — on the developer's machine, in their working tree. Provider keys are encrypted at rest, and spend caps are hard per key and per team: an over-cap request is refused before any inference is bought, so a runaway agent cannot become a runaway bill.

Security posture

• Source never leaves your environment. Tools run locally; the gateway handles accounts, keys, caps, and routing — not files.
• Prompts transit only to your provider, on your key, under your provider agreement.
• No training on customer code. Ever.
• Zero markup. Inference bills to your provider account at provider prices.

Start with a concierge pilot

The fastest start is a hosted concierge pilot: one team, their own provider key, individual accounts, and a short, measured run. The pilot is judged on two numbers — cost per resolved task and task success rate, both read from per-key usage — so the decision to roll out across the VPC is made on evidence, not demos. The full VPC deployment follows once the pilot proves out.