Service Level Agreement

Uptime is a number.
Response is a clock.

We publish the targets we hold ourselves to and the credits we owe when we miss. Numbers below apply to paid tiers; Solo runs best-effort but the same status page shows real availability.

Targets

What we commit to per tier.

Solo is best-effort and unmetered. Studio, Team, and Enterprise carry SLA targets with service credits if we miss. Targets measured monthly, calendar-month boundaries, UTC.

CommitmentTierSoloTierStudioTierTeamTierEnterprise
Monthly uptime targetBest-effort99.5%99.9%99.95%
Maximum monthly downtime budget~3h 39m~43m~21m
P0 acknowledgeBest-effort8h biz4h biz, 24/7 weekend on Team+1h, 24/7
P1 acknowledgeBest-effort24h biz8h biz4h, 24/7
Service credit on miss5% MRR per 1pp miss, capped 50%10% MRR per 1pp miss, capped 100%Custom (contract)
Status page✓ + dedicated subscription
PostmortemPublic summaryPublic summaryCustomer-private summaryJoint review meeting

Severity

What counts as P0, P1, P2.

P0 — Outage
Hard down.
Editor is unreachable. Saved sites won't load. Sandbox jobs fail to start. Affects all paying customers OR a single Enterprise account globally. Pages on-call within target window, public incident on status page within 15 minutes of confirmation.
P1 — Degradation
Bent, not broken.
Latency more than doubled, or a load-bearing surface (billing portal, marketplace publish, vibe-edit) is intermittent. Workarounds may exist. Tracked publicly on the status page; postmortem within 5 business days.
P2 — Anomaly
Cosmetic or workaround-able.
Cosmetic, single-tenant, or workflow-around-able. Triaged inside the planning system; surfaced to the affected operator over their support channel; resolved within next sprint.

Operations

Why we can promise this.

Continuous deploy + canary — Every change to the platform deploys via Vercel preview, runs the eval suite against a representative project, and rolls out to production behind a 5% canary. Rollbacks are atomic and one click.

Sandbox isolation — Vibe-edit jobs run in Vercel Sandbox — Firecracker microVMs with hard CPU + memory + duration ceilings. A runaway agent in one tenant cannot starve another tenant's jobs.

Audit + replay — Every API call lands in our planning telemetry stream. Replays are deterministic; postmortems start from real traces, not memory.

Credits

When we miss, here's what you get.

Service credits apply automatically to the next billing cycle's MRR when monthly uptime falls below the contracted target. Credits are calculated per the table above (5% per percentage point on Studio, 10% on Team, custom on Enterprise). Credits cap at the listed maximum unless your contract specifies otherwise.

To claim a credit you didn't see auto-applied, email benjamin@gad.tools within 30 days of the incident's resolution date. We respond within 5 business days with the verdict and the dollar figure. There's no form. There's no portal.

Last updated: 2026-05-02 · Effective for all subscriptions taken on or after this date.