CORE://NEO
Our own agent framework, built from zero
A proprietary security-first AI agent framework — command blocklists, encrypted secret store, prompt-injection defense — built to be owned, not rented.
In the lab — active R&D
The problem
Every business building on AI agents is renting its core capability from a platform. The stack that runs your operation should be an asset you own — and most agent frameworks bolt security on last.
The system
NEO is BTD's from-scratch agent framework, designed security-first: layered command blocklisting, a ChaCha20-encrypted secret store, and prompt-injection defense baked into the architecture — patterns distilled from studying nine open-source agent frameworks. The roadmap: internal daily-use prototype, then client delivery vehicle, then product.
How it flows
- 01
Instruction in
Slack, dashboard, or scheduled job
- 02
Security layer
Blocklist + injection defense screens every action
- 03
Agent plans
Tool selection against encrypted secret store
- 04
Execute & log
Auditable action trail
Who this is for
Today: BTD's own R&D edge and the security thinking behind every client build. Tomorrow: the delivery framework client systems ship on.
FOOTAGE INCOMING
System walkthrough in production
Want a system like this?
Or get the weekly breakdown of how systems like this get built: