MemMesh v0.1.2

Local semantic search + consolidation, prebuilt installers (macOS/Linux/Windows). Binaries attached by CI.

MemMesh v0.1.1

Patch release. Fixes a silent data-loss bug in observe and aligns the plugin with what the open-source engine actually does.

Fixes

• observe no longer silently drops input. The heuristic extractor only matched first-person phrasings (I prefer…, we decided…) and discarded everything else — third-person statements, emails, most prose. Substantive text is now persisted verbatim as a raw observation when no structured rule matches, so nothing you tell the engine is lost. (Questions, filler, and code are still filtered.)
• Three CRUD MCP tools added — memory_delete, memory_supersede, memory_stats (MCP surface 7 → 10).
• Honest skill gating — prediction / behavior / graph-reasoning skills now clearly indicate they need hosted mode and degrade to search/recall on a local install, instead of calling a tool that isn't there.
• Plugin & marketplace descriptions retuned to lead with the local-first memory store.

Upgrade

Rebuild from source (prebuilt binaries land in a follow-up release). Anyone on v0.1.0 should upgrade — observe was effectively a no-op for most inputs before this.

Full changelog: v0.1.0...v0.1.1

MemMesh v0.1.0

Persistent, self-improving memory for AI agents — a local-first memory engine in Rust. One binary that runs as a desktop component (SQLite) or a server backend (Postgres), speaks the Model Context Protocol, and wires into your AI tools with a single command.

Highlights

• Uncapped and open — Apache-2.0, no memory limits in the self-hosted engine.
• Local-first — your data never leaves your machine; no external vector DB or search cluster required.
• Typed, scoped, bi-temporal memory — facts, contacts, and relationships with valid_from/valid_to and provenance; records when something happened in the world vs when it was stored.
• MCP-native — observe, save, recall, search, and more, exposed as MCP tools for Claude Code, Cursor, and any MCP client.
• Zero-LLM knowledge graph — entities and edges built deterministically, no per-memory LLM calls.
• Sync — push/pull between local and server.

Install

See the README for install options. Source archives are attached below.

Prebuilt binaries (macOS / Linux / Windows) and a Homebrew tap are coming in a follow-up release.