WaveMind launch: dynamic memory for agents and apps

[CaspianG]

WaveMind is now ready for broader developer feedback.

The short version: WaveMind is a local-first dynamic memory layer for agents and applications. It keeps durable memory in SQLite or Postgres, uses vector search for candidates, and then applies memory state: hotness, priority, decay, TTL, namespaces, tags, audit events, and optional graph dynamics.

This is not meant to replace Chroma, Qdrant, Pinecone, or Postgres. The goal is different: make memory behavior reusable on top of ordinary retrieval.

Try it

python -m pip install wavemind
wavemind remember "The user prefers short answers" --namespace demo
wavemind query "answer style" --namespace demo

From a clone:

python examples/demo.py
python examples/dynamic_memory_demo.py

The dynamic demo shows:

• corrected newer facts outranking stale facts;
• temporary memory expiring;
• namespace isolation;
• index-health reporting.

Where feedback would help most

1. Memory model: hotness, decay, TTL, corrections, graph dynamics.
2. API ergonomics: Python, CLI, FastAPI, LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen.
3. Benchmarks: LoCoMo, LongMemEval, BEIR/SciFact, dynamic memory policy, ANN curves.
4. Production gaps: FAISS/Qdrant/pgvector profiles, observability, backups, deployment docs.

Honest limitations

Static vector search can be faster. WaveMind is currently strongest where memory policy matters: stale suppression, corrections, TTL, namespaces, auditability, and recall state. Full answer-quality leaderboard results are still future work.

Useful links:

• README: https://github.com/CaspianG/wavemind#readme
• PyPI: https://pypi.org/project/wavemind/
• Use cases: https://github.com/CaspianG/wavemind/blob/main/docs/USE_CASES.md
• Launch kit: https://github.com/CaspianG/wavemind/blob/main/docs/LAUNCH_KIT.md
• Roadmap: https://github.com/CaspianG/wavemind/blob/main/docs/ROADMAP.md