v0.1.0 — mneme v1 core

Baseline release marking the v1 core: a standalone Go library giving agents persistent, searchable long-term memory.

go get github.com/AccursedGalaxy/mneme@v0.1.0

Added

• Public Memory API — Add, Search, Get, Delete, Close — with functional options (New(...Option)) and MNEME_* environment wiring.
• Additive ingestion pipeline: retrieve existing → integer-relabel → one LLM extraction → defensive parse → hash dedup → embed → persist.
• store/sqlite: pure-Go (modernc) backend, float32 BLOB embeddings, cosine similarity computed in Go, per-scope isolation.
• Providers: SDK-free OpenAI-compatible HTTP client (provider/openai) and a deterministic offline provider/fake (scripted LLM + feature-hashing embedder) for tests.
• Versioned extraction prompt (extractionPromptV1), owned in-repo.
• Eval harness (cmd/eval, eval/): 18 fixtures, LLM-judge and deterministic metrics, recording a baseline aggregate of 0.97 on gpt-4o-mini with text-embedding-3-small.

---

This is the first tagged baseline. v2 levers (consolidation, rerank/multi-query, temporal grounding) accumulate toward v0.2.0 — see CHANGELOG.md.