Releases·Tool Release
Engram: local-first memory for coding agents. SQLite FTS5 + BM25, optional local embeddings, no network at recall time (MIT)
A developer released Engram, a free, MIT-licensed local-first memory system for coding agents. It stores notes as plain Markdown on disk and indexes them with SQLite FTS5 and BM25, with no network access at recall time. It targets developers wanting agent memory that stays fully local and disposable.
11 engagement1 sourceSat, Oct 3, 2026, 01:28 PM
AI Overview
- Author says recall uses BM25 over a rebuildable SQLite FTS5 index built from plain Markdown files, fusing cosine results only if a local embedding model exists
- Author warns recall@5 of ≥90% across 20 seeded queries is a regression gate in the test suite, not a benchmark
Entities
Related
Products·launched
Synapse: Markdown + SQLite memory store for LLM agents with MCP retrieval
Synapse is an open-source, local memory store for AI agents that keeps history and wiki pages as Markdown files and builds a disposable SQLite FTS5 index. Agents access it via MCP (search and propose_note), retrieving only relevant pages instead of ingesting the whole archive. It solves the problem of inspectable, durable agent memory without depending on a hosted store.
Synapse is an open-source, local memory store for AI agents that keeps history and wiki pages as Markdown files and builds a disposable SQLite FTS5 index. Agents access it via MCP (search and propose_note), retrieving only relevant pages instead of ingesting the whole archive. It solves the problem of inspectable, durable agent memory without depending on a hosted store.
12 engagement1 sourceredditWed, Sep 30, 2026, 03:21 PM
Releases·Tool Release
Developer releases Memory Engine MCP, a local-first graph-aware memory server for Claude, Cursor, Cline
A developer open-sourced Memory Engine MCP, a local-first long-term memory system for AI coding tools that models information as typed atoms connected by typed bonds. It uses a hybrid ranking pipeline combining SQLite FTS5 full-text search and semantic similarity via Ollama, aiming to connect and curate data rather than just store it.
A developer open-sourced Memory Engine MCP, a local-first long-term memory system for AI coding tools that models information as typed atoms connected by typed bonds. It uses a hybrid ranking pipeline combining SQLite FTS5 full-text search and semantic similarity via Ollama, aiming to connect and curate data rather than just store it.
3 engagement1 sourceredditWed, Jul 29, 2026, 09:08 AM
Products·launched
Dejavu: local, repo-scoped memory for coding agents
Dejavu is a memory layer for coding agents that persists session context in a local SQLite file, scoped to the repository. It uses FTS5 full-text search to store handoffs and decisions, eliminating the need for cloud dependencies or embeddings. The tool solves the problem of agents forgetting context between sessions, saving developers from re-explaining past decisions.
Dejavu is a memory layer for coding agents that persists session context in a local SQLite file, scoped to the repository. It uses FTS5 full-text search to store handoffs and decisions, eliminating the need for cloud dependencies or embeddings. The tool solves the problem of agents forgetting context between sessions, saving developers from re-explaining past decisions.
5 engagement1 sourceredditThu, Jul 30, 2026, 02:58 PM
Products·launched
Llmem: Local persistent memory for AI coding tools via MCP/REST
Llmem is a lightweight, local persistent memory system for AI coding assistants. It stores project learnings and preferences in a SQLite file, exposed via MCP and REST APIs, so AI tools can remember past mistakes and patterns without embeddings or markdown files.
Llmem is a lightweight, local persistent memory system for AI coding assistants. It stores project learnings and preferences in a SQLite file, exposed via MCP and REST APIs, so AI tools can remember past mistakes and patterns without embeddings or markdown files.
1 engagement1 sourcehackernewsFri, Aug 7, 2026, 08:40 AM
Products·launched
Tennis: local/offline CLI context store for chats and files
Tennis is a CLI that indexes and searches files, Claude/Codex chat sessions, and other context entirely locally using SQLite FTS5 keyword search plus vector embeddings. It addresses the problem of LLM chat history being siloed in separate apps, for users who want searchable context without cloud subscriptions or server-side storage.
Tennis is a CLI that indexes and searches files, Claude/Codex chat sessions, and other context entirely locally using SQLite FTS5 keyword search plus vector embeddings. It addresses the problem of LLM chat history being siloed in separate apps, for users who want searchable context without cloud subscriptions or server-side storage.
2 engagement1 sourcehackernewsFri, Oct 16, 2026, 03:17 PM