Everyone is renting dragons. The big labs charge you per token for models you’ll never own. I went the other way and trained my own: • Cast — 3.3M params that turns plain English into database queries • Imagine — 1.6B full fine-tune. 40/40 on held-out tests

@boardyai what’s on the agenda? Looking for builders to integrate NEDB-engine, and let’s talk Postgres SQL

Provenance should be a rule the database enforces, not something you bolt on after the audit. @boardyai introduced me to Najeed Khan, founder of AgentV, last week. About ten minutes in we were circling the same problem from opposite ends. Agentv stress-tests autonomous agents in simulation before they reach production. Outcomes, permissions, tool calls, state transitions, all replayed and checked, with cryptographically signed evidence at the end. They are agnostic about the storage, but are validated on Postgres today. A signature proves a result was not changed after it was recorded. When a number turns out wrong, the question in the room is what the data looked like on the day the agent produced it, and which input caused it. If the store only knows the current state, that question has no answer, and no amount of logging added later recovers it. That is why I built neDB the way I did. Every write sits in a BLAKE2b hash chain. Every record carries a caused-by pointer back to the input that produced it. Deleting something leaves a tombstone, so history never gets garbage collected. The store is bitemporal, so you can ask what the data was on any date instead of only what it is now. None of that is optional per table. It is how the engine writes. We spent the most time on the migration question, because for a Postgres shop the answer is you don't migrate. Import the wrapped Postgres client, wrap the connection, turn on shadow writes. The app keeps speaking Postgres while every write also lands in neDB with its history intact. Nothing is decided. He is reading the docs and the repo and sending me questions, and I'll let Najeed take the verification side from here. If you run agents in production and one of them gets something wrong, can you show what the data looked like that day, or only what it is now?