Deterministic Continuity for AI Systems
AI systems lose consistency over time.
- Decisions are forgotten
- Completed tasks remain open
- Contradictions appear across sessions
Chat history is not reliable memory.
Memory OS detects and repairs inconsistencies.
It verifies:
- decisions vs actions
- committed state vs open loops
- policies vs current context
Then repairs mismatches deterministically.
- Decision: Authentication is deferred to v1
- Action: Build authentication system
→ contradiction
- Policy: All APIs must include request_id
- Open loop: Implement request_id
→ stale loop
- detects inconsistency
- explains the issue
- suggests fix
- repairs if possible
proposal → approval → commit
No direct AI writes.
LLM-based memory is not reliable for system-level consistency.
Memory OS is:
- deterministic
- explainable
- reproducible
- continuity verification engine
- repair system for AI context
- chat history storage
- vector database
- RAG system
pnpm install
pnpm build
createdb memory_os
node infra/db/migrations/run.js
node apps/worker/dist/worker.js ingest-raw --file=data/fixtures/raw/sample-session.json
node apps/worker/dist/worker.js rebuild-context-cache --project=proj_memory_os
pnpm --filter @memory-os/evals run eval:run
node apps/worker/dist/worker.js rebuild-context-cache --repair- Structure first.
- Determinism over probability.
- Continuity over convenience.
- MIT
Masaaki Sakamoto
https://masaaki.ai
Building systems for Human × AI continuity.