Skip to page content
Proof Docs

Cost controls

Proof starts with live work and AI turned off. Enable them separately, set provider limits, then run one saved test.

These settings apply to a Proof deployment. The first-test procedure below uses the current website workflow.

Safe defaults

Local setup, linting, type checks, unit tests, builds, Workflow checks, CLI builds, and the documentation site do not create Vercel resources or call an AI provider.

dotenv
PROOF_LIVE_EXECUTION_ENABLED=false
PROOF_AI_EXECUTION_ENABLED=false
PROOF_MODEL_TIER=budget
PROOF_ALLOW_PREMIUM_MODELS=false

Independent interlocks

VariablePurposeDefault
PROOF_LIVE_EXECUTION_ENABLEDAllows live Workflows and isolated workspaces.false
PROOF_AI_EXECUTION_ENABLEDAllows any AI SDK or Gateway request.false
PROOF_MODEL_TIERSelects budget or premium model policy.budget
PROOF_ALLOW_PREMIUM_MODELSSecond opt-in required for premium routing.false
PROOF_ALLOW_LIVE_VERIFICATIONAllows billable integration verification scripts.false
PROOF_ALLOW_BILLABLE_SETUPAllows creation of a reusable browser snapshot.false
PROOF_ALLOW_UNCACHED_BROWSER_RUNTIMEAllows a fresh Chromium download instead of a snapshot.false
PROOF_MONITORING_ENABLEDControls scheduled monitoring. A separate safety stop currently blocks execution.false
PROOF_MONITOR_AI_ESCALATION_ENABLEDAllows AI to recheck repeated monitor failures.false
PROOF_MODEL_MAX_RETRIESCaps AI SDK retries for each logical model call; run telemetry records the resulting attempts.0
PROOF_HARD_SPEND_CAP_CENTSCeiling for reported Gateway spend plus active reservations. Values above 5000 are rejected. Zero pauses further paid work.1000
PROOF_MONITOR_MIN_CADENCE_MINUTESRejects schedules faster than the provider Cron wake-up.1440

Proof checks reported AI Gateway spend and active reservations before admitting new paid work. The default limit is $10, and a deployment can raise it to $50. A Proof platform admin can save a shared limit in Spending controls. The lower of both limits applies. This is not a complete provider invoice meter or a promise that all charges reset monthly. Operators must also configure provider spending limits.

Run the first test

  1. Use a capped test database and apply reviewed migrations.
  2. Set up one Clerk development organization and one fixture website.
  3. Set provider spending limits before enabling paid work.
  4. Enable live execution and AI for one bounded Mapping session on the fixture website.
  5. Review the proposed cases and exact expected text. Confirm one test with a small action scope, then wait for writing and fresh browser verification to pass.
  6. Review its evidence, model usage, isolated workspace cleanup, and database use before more work.
  7. Turn AI off while live execution stays on. Run that verified saved test again and inspect its result. This Run uses no model calls.

Operate with AI paused

Mapping and writing require AI. Saved mapping tests can run with AI off, if live execution remains enabled. A saved Run does not call a model or start automatic repair. Review failed Run evidence and request a repair when AI is available.

Stop conditions

Turn the settings off at once if a Job loops, a deployment does not become ready, isolated-workspace cleanup fails, usage exceeds the test limit, a provider ignores its limit, or a delivery repeats. Keep the Job record. Check the expected retry cost before you try again.