INSTNCT T1 Reflex Engine
This preview introduces the engine contract: local-first execution target, Exact Mode refusal, visible limits, and a Proof Pack before runnable claims.
GitHub
Founder-built. AI-assisted. Evidence-gated.
VRAXION is the solo-built, AI-assisted project behind INSTNCT: a local reflex engine for known paths. It is being built to move fast where support exists, refuse where it does not, and ship proof before big claims.
Public path
The INSTNCT T1 Reflex Engine is the first concrete target: a local reasoning engine that chooses a supported path, returns a bounded result, or stops cleanly when the path is not supported.
This preview introduces the engine contract: local-first execution target, Exact Mode refusal, visible limits, and a Proof Pack before runnable claims.
Training data research
This path is not a model announcement. AnchorCell is a structured training-data research direction with a v2 authoring schema baseline: one decision cell, explicit trust separation, grounded facts, adversarial branches, and public-safe export views.
The goal is a reviewable example format that makes shortcuts, uncertainty, evidence, and export rules visible before any training artifact is trusted.
What VRAXION is
This is the VRAXION AI/reasoning surface in public: not a chatbot identity, but a named engine with modes, paths, and verification gates.
The engine target starts by selecting a known path. If no supported path exists, the answer is not forced.
Exact Mode is the default contract: answer only inside supported paths, otherwise refuse cleanly.
The runnable claim is defined by a signed artifact, checksums, notes, and local verification steps.
Evidence surface
Public today is the INSTNCT preview, SDK/docs, release tag, status notes, and capability scope. The offer is simple: a local reflex reasoning engine that says yes inside known paths and no outside them. The runnable T1 Proof Pack still has to ship.