zer0dev is a trust layer for IDE agents. It hosts boot, skills, kernel law, and an Audit Done Gate over HTTP so sessions can require observations before work is called done. It does not replace Cursor, Claude Code, or Copilot.

Students who want governed IDE sessions. New accounts need a school email that ends in .edu. If you only need a chat model with no governance loop, you do not need zer0dev.

Does zer0dev upload my source code?

No. Project memory stays on your machine under a configured memory path. The host stores account and usage metadata, not your repository.

How do agents authenticate?

Create an install key (zd_live_…) after sign-in and send it as Authorization: Bearer. Agents can discover OAuth metadata at /.well-known/oauth-authorization-server and exchange the key at POST /api/oauth/token.

How much does zer0dev cost?

Free for students with a .edu school email. There are no paid plans or subscriptions. Non-students who want to test can contact the author on LinkedIn.

Does zer0dev predict whether an agent's change is safe?

No. The model predicts tokens. Consequences sit outside the model. zer0dev is a session governor: it can require non-model oracle receipts before work is called done. Those receipts are observations of named checks, reported by the agent. The host does not run your compiler or browser, and it does not claim remaining consequences are known.

Where should an agent start?

Read /llms.txt for when-to-use guidance, /api/openapi for the route catalog, and /developers for human docs. Prefer Accept: text/markdown on content routes.

zer0dev Developer API

Quickstart

Sign in. New accounts need a school email that ends in .edu.
Open Boot and paste Install into your IDE.
Create an install key in Settings (zd_live_…). Send Authorization: Bearer.

Public endpoints

GET /api/public/discoveryMachine-readable discovery hub
GET /api/public/onboardingFree tier, sandbox, install keys
GET /api/public/sandboxVerified /boot reachability
GET /api/public/capabilitiesLive endpoint probe
GET /api/public/developer-indexResource catalog
GET /api/healthLiveness

OpenAPI and indexes

Markdown twins

Send Accept: text/markdown on HTML pages or fetch twins directly.

Auth for agents

OAuth metadata: /.well-known/oauth-authorization-server

Token: POST https://zer0dev.vercel.app/api/oauth/token with grant_type=client_credentials. Scopes: content:read, audit:write, telemetry:write, memory:write, mcp:connect, install:manage.

Content API

All content routes require Bearer (install key). Start at GET /api/boot for the session index — module ids are not listed on public pages without auth.

GET /api/bootSession index (Bearer required)
GET /api/kernel/{module}Kernel modules
GET /api/skills/{id}Skills catalog and bodies
POST /api/quality/auditAudit plus verdict and receipts

MCP

https://zer0dev.vercel.app/api/mcp — Audit and telemetry (Bearer)

FAQ

What is zer0dev?: zer0dev is a trust layer for IDE agents. It hosts boot, skills, kernel law, and an Audit Done Gate over HTTP so sessions can require observations before work is called done. It does not replace Cursor, Claude Code, or Copilot.
Who is zer0dev for?: Students who want governed IDE sessions. New accounts need a school email that ends in .edu. If you only need a chat model with no governance loop, you do not need zer0dev.
Does zer0dev upload my source code?: No. Project memory stays on your machine under a configured memory path. The host stores account and usage metadata, not your repository.
How do agents authenticate?: Create an install key (zd_live_…) after sign-in and send it as Authorization: Bearer. Agents can discover OAuth metadata at /.well-known/oauth-authorization-server and exchange the key at POST /api/oauth/token.
How much does zer0dev cost?: Free for students with a .edu school email. There are no paid plans or subscriptions. Non-students who want to test can contact the author on LinkedIn.
Does zer0dev predict whether an agent's change is safe?: No. The model predicts tokens. Consequences sit outside the model. zer0dev is a session governor: it can require non-model oracle receipts before work is called done. Those receipts are observations of named checks, reported by the agent. The host does not run your compiler or browser, and it does not claim remaining consequences are known.
Where should an agent start?: Read /llms.txt for when-to-use guidance, /api/openapi for the route catalog, and /developers for human docs. Prefer Accept: text/markdown on content routes.

zer0dev developer API