🤖 infrastructure for autonomous ai agents
🧠 memory • 🔒 privacy • ⚡ local inference
agents can plan and call tools. the missing piece is infrastructure that stays on the machine: memory you can prove, tool calls you can check, and text you can redact without sending it to a vendor.
small libraries meant to sit on a hot path:
🧠 provable memory — show a fact was stored, or that it never was, without handing over the store
🔒 privacy by default — zkSNARK proofs • confidential compute • no surveillance
⚡ local execution — checks run in-process • no round trip • built for the hot path
✅ tool-call integrity — prove a call matched its schema and a replay matches the log
tech stack:
fast rust libraries developers run in-process • proofs, checks, and redaction on the hot path
tech: rust • sparse merkle trees • schema proofs • local inference
compute on encrypted data • generate zero-knowledge proofs • optimized 30s → 8s
tech: zkSNARKs (groth16, plonk) • FHE (bfv, ckks) • distributed systems
provable agent memory → was this fact stored, and can a third party check it?
programmable privacy → compute and disclose only what a proof allows
local inference → checks that finish in-process, not in a hosted api
tool-call integrity → a call matched its schema, and a replay matches the log
