Brain AI — grounded reasoning across the GlueArrow graph.
The reasoning surface of the ecosystem is live. Brain reads the same ledger as Command Center and answers in the same voice — with evidence, not guesses.
3 min readBy The GlueArrow Editors, Newsroom
Brain AI is the reasoning surface of GlueArrow. It is live today with a six-tab admin dashboard and ten API endpoints, reading from more than sixteen live data tables. Brain answers questions about the network the same way the network speaks to the regulator — with a row of evidence behind every line.
What is in the first release
- A schema registry that gives Brain a typed map of what the platform actually holds.
- A safe-SQL layer — sixteen wrappers for the most common lookups, so the model never invents a column.
- A two-pass reasoning loop — introspect first, answer second — with logs the meta-review can audit.
- A meta-review Celery task that nightly grades conversations and queues findings for human approval before anything is promoted.
What Brain is not
Brain is not a chatbot dressed up as an analyst. It does not freestyle; it reads and replies. If the graph does not have the answer, Brain says so. If the user is not authorised to see the answer, Brain does not see it either.