macOS AI IDE built on a deterministic engine (Vera-a): typed refusals instead of guesses, order-invariant answers, and a covenant guard that catches an LLM drifting from what you told it. Offline, no embeddings. Every claim is backed by a pre-registered measurement you can re-run in 45s (89 forks + 50 checks).
refactoring security metal gatekeeper quantization thunderbolt claude zero-trust knowledge-engine local-llm deepseek ai-coding macos-ide jgen
-
Updated
Sep 27, 2026 - Python