lbrain.ai

Three cookbooks.
None of them are "Alice likes Slack."

Self-seed a toy lair from the cookbooks in lbrain-coding-agents. Five minutes, no account, no API key.

1 · Abstain

The record does not support an answer.

The demo corpus has a timeout for the metrics exporter. It does not have a timeout for the ingest API. A neighbour that looks like an answer is a near-miss. Serving it as the timeout is RAG. Abstaining is LBrain.

$ lbrain query "what is the request timeout for the ingest API?"

2 · SUPERSEDED

Buried is not forgotten.

September said rollback was --force. March replaced it with --safe after events dropped. Query for the current flag; search still finds the old one, marked SUPERSEDED. The old flag must not authorize an action.

$ lbrain query "what is the rollback flag for the staging deploy?"
$ lbrain search "rollback --force"

3 · Model swap

When the model changes, the mind does not start over.

Record a decision on Claude Code. Ask the same LBRAIN_HOME from Codex or Grok Build in a fresh session. If the second model cites source and date, that is the demo. If it invents a flag, check lbrain whoami in both sessions. This is a positioning demo you can run, not a published measurement of model-swap survival. Say that honestly.

Installer wires both mouths to one brain.