NoDrift keeps AI work moving in order: start clear, check the answer, approve what matters, and continue without starting over.
NoDrift keeps each AI work session tied to source truth, approval, heartbeats, corrections, branches, evidence, and carryover.
Name the real task, current goal, and what matters before the AI starts moving.
Keep today's work from quietly expanding into side quests or future features.
Use project files and records as proof. Chat memory is not the authority.
Use programmed refreshes and behavioral checks to route risk back into governance.
Separate confident AI output from verified fact before relying on it.
Important moves wait for the user's exact yes before the work changes.
Do the approved work without widening the task or changing direction.
Check what actually changed, not just what the AI says changed.
Name wrong assumptions and preserve the fix so the same error stops returning.
Give separate objectives, risks, or decision paths their own record.
Keep proof, limits, and open questions connected to the claim.
Let the next session resume from the record instead of restarting.
NoDrift maps trigger points like startup, phase change, readiness, public/private boundaries, external action, package work, payment, compaction, source conflict, correction, token burn-rate, and website claims to the right governance refresh.
Behavioral Heartbeats respond when the user calls out drift, over-effort, wordiness, stubbornness, unsupported source claims, or governance-operation problems. The assistant may also initiate the same check as a self-audit.
Heartbeats do not control the model or grant permission. They route the session back to correction, source truth, approval boundaries, and the next safe action.