Kernel · online

An AI that checksits own work.

BrainOS is a Python kernel that keeps agents honest. It names what counts as proof before the work starts, ties every claim to a source, and refuses to close a task that doesn’t check out. On my machine: SessionStart pushes standing rules into Claude Code; an MCP server (recall, learn, standing context, metrics) makes the same surface callable from other clients. The animation below is the loop — not the whole CLI.

brainos · loop

idle

trace

readyawaiting task

one loop · MCP callable

CLI commands
cover the loop

claims closed
without a source

every

Claude session
hook fires

fail-open

errors never block
empty inject, exit 0

What you just watched.

An illustrated walk through the loop. The real kernel runs as the brainos CLI on your machine; this animation shows the eight steps a single task takes through it. No claim survives without a source it can point to, and the loop only closes when the proof bar it set up front is met.

scope

Name the task and what done looks like.

route

Pick where the answer should come from.

claim

State exactly what's being asserted.

source

Bind the claim to the origin it came from.

check

Test the claim against that source.

gate

Does it clear the proof bar set in step 1?

witness

Stamp a receipt anyone can re-verify.

Close — only if the work checks out.

What BrainOS does today.

Plain facts, all currently true:

· Python kernel at C:\tmp\brainos-kernel · CLI exposed as brainos
· Wired into Claude Code via a global SessionStart hook in ~/.claude/settings.json — fires on every session, every directory, since 2026-05-26
· Per-project context injection: resolves cwd against profiles/*.json · match → injects North Star, Inherited Rules, Forbidden Actions
· Inherited Rules are proof-gated: recorded via brainos learn, only kept when a check confirms they helped
· Fail-open contract: any error → emits nothing, exits 0, never blocks a session
· CLI surface: classify · plan · ingest · claim · propose · close · query · brief