OpenNeko Blog
Notes on building and forward deploying OpenNeko.
Your AI's memory can be poisoned. OpenNeko keeps receipts.
HMAC-sealed memories, TTLs on learned facts, an invisible git repo per workspace, and a hash-chained audit log — the actual mechanisms, with code.
Read post →Stop pasting your database schema into the prompt
Four mechanisms — a prefetched knowledge pack, role-shaped warm starts, a bounded self-correction loop, and lossless columnar compaction — that get an agent to hand-tuned speed on a database it has never seen.
Read post →Your API key never enters the box. Grep it and check.
The threat model, the sandbox architecture, and the actual code behind running an LLM agent as untrusted code — with a three-line verification you can run yourself.
Read post →Rent the intelligence. Own the context.
Why your business context should live with you, not inside someone else's model.
Read post →