Publishing tools
Ship an AI tool with zero AI cost to you. Submit it, pass the automated check, and it's in the store.
The three shapes
- Tools & apps — from a tiny declarative tool (a manifest + a prompt, no code) to a full app. Local‑first: runs on the user's own device model where it can, in Stevai's sealed sandbox.
- Connectors — let the assistant read and act in a system your users already have (email, CRM, your own SaaS). See .
- Embedded backend — build your own product on Stevai's AI and earn a share. Uses the with an API key.
How a submission flows
Every submission (and every update) runs through the same pipeline:
- Automated precheck — validates your manifest, checks policy (no subscriptions), reconciles what your code does against what it declares, and assigns a risk tier. It returns a report split into blockers (must fix) and needs‑human‑eyes risk factors, plus a fixed fee quote if a human review is needed.
- Auto‑publish — a contained, local‑first, clean tool from a verified developer publishes immediately (a kill‑switch and monitoring stay on).
- Human validation — anything higher‑risk (native desktop, connectors, requested network/filesystem/device access, sensitive categories) is reviewed by a certified validator. You get a fixed fee and time estimate up front and pay only when you approve.
Submit and see your report at stevai.com/app/publish.
The rules
- Buy once, own it. Tools are a one‑time purchase; you may charge for major updates. No subscriptions for tools.
- Declare everything. A capability your code uses but your manifest didn't declare is an automatic blocker. See the .
- Security split. You own your tool's correctness and security; Stevai owns containment (the sandbox), policing (instant kill‑switch), and store compliance.
Updates
Ship an update by resubmitting. The check diffs against the last reviewed version, so a small, in‑scope fix is re‑reviewed cheaply — but any change outsidethe flagged area is caught and re‑reviewed, so you can't smuggle a change into a small fix.