Start building
Kernall documentation
Set up the safety-gate preview, connect your agent, and decide what happens before an action runs.
Run locally and get a real policy decision.
02 / REFERENCEExplore the API ↗Requests, responses, authentication, and limits.
What is Kernall?
Kernall evaluates an action before your agent performs it. Your integration submits the proposal, Kernall checks your project policy, and the API returns approved, denied, or requires_human. Each check is recorded so you can inspect its decision.
The current implementation is an API safety gate. Your integration must call it and honor its decisions. The broader isolated sandbox runtime is under development.
What you can use today
| Surface | Current status | Guide |
|---|---|---|
| HTTP safety gate | Implemented in the developer preview | Action checks → |
| Policy, keys, reviews, logs | Backend APIs implemented | Policy → |
| Python client | Available in the source repository | Python SDK → |
| Web dashboard | Demo UI; server integration incomplete | Limitations → |
| Desktop, packaged CLI, MCP | Planned; no released installers or server | Availability → |
Find what you need
- New to Kernall? Follow the local quickstart, then read core concepts.
- Connecting an agent? Read action checks, human approvals, and the Python client.
- Managing access? Configure policy, manage API keys, and inspect audit records.
- Looking for the Mac app? See desktop availability and the planned installation flow.
- Something failed? Start with troubleshooting.
How to read these docs
Instructions describe behavior in the current source. Planned features have explicit availability notes. Example IDs and keys are placeholders; use values returned by your own instance.
The preview has no committed release date or compatibility guarantee. Read data and security before deciding which workloads to connect.