This roadmap is a design target for community planning. It is not a release promise.
- Publish cleaned English and Chinese whitepapers.
- Add contribution, security, governance, and third-party policies.
- Add vetted reference repositories as pinned submodules.
- Open GitHub Issues and Discussions for architecture review.
- Define benchmark adapters for LongMemEval, LoCoMo, AppWorld, and selected poisoning tests.
- Record reproducible baseline procedures before implementation claims are made.
- Define acceptance thresholds as targets, not marketing claims.
- Draft PostgreSQL schemas for chat logs, candidate memory, trusted memory, playbook entries, provenance chain, reflection logs, and audit events.
- Decide which schema fields are stable public interfaces.
- Prototype embedding reindex behavior as a rebuildable background task.
- Implement inbound scanner interfaces.
- Implement quarantine state transitions.
- Add provenance signing and verification.
- Add audit-event logging before enabling any trusted-memory promotion.
- Prototype local consolidation jobs.
- Evaluate local model choices against privacy, cost, latency, and quality.
- Add learned-context and anticipatory-cache write paths through the Shield.
- Implement retrieval prefetch and tool-driven deep search.
- Implement ACE-style playbook storage and intent-scoped injection.
- Measure interaction latency and memory quality regressions.
- Run full benchmark and adversarial suites.
- Document migration and backup strategy.
- Publish release notes with methodology and known limitations.