I'm working on MemTether - a cross-client AI memory hub where heterogeneous AI agents share the same physical SQLite database via file-level pointers.

No server. No sync. Zero API cost.

Multiple AI clients (Codex, WorkBuddy, OpenClaw, ZCode, etc.) all point to the same memory.db file — so when you switch from one client to another, your memory comes with you.

• File-level pointers (junction/symlink) — not sync, not copies
• Dual-timeline governance — business time + recorded time, supersession version chains
• Hybrid retrieval — vector (bge-m3) + FTS5 BM25 + keyword + literal, RRF fusion + cross-encoder reranking
• Conflict detection — automatic detection + human review, not blind overwrite
• Memory Exchange — governance semantics travel with memory across systems (5 adapters)
• Zero cost — 100% local, offline capable, zero API fees

\�ash pip install memtether \

• GitHub: MemTether/MemTether
• PyPI: memtether
• License: Apache-2.0