# claude-mem-lite

> Lightweight persistent memory system for Claude Code — FTS5 search, episode batching, error-triggered recall

- **URL**: https://www.freshcrate.ai/projects/claude-mem-lite
- **Author**: sdsrss
- **Category**: MCP Servers
- **Latest version**: `v2.89.0` (2026-06-02)
- **License**: MIT
- **Source**: https://github.com/sdsrss/claude-mem-lite
- **Language**: JavaScript
- **GitHub**: 37 stars, 1 forks
- **Registry**: github
- **Tags**: `claude-code`, `fts5`, `hooks`, `javascript`, `mcp`, `memory`, `persistence`, `sqlite`

## Description

Lightweight persistent memory system for Claude Code — FTS5 search, episode batching, error-triggered recall

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v2.89.0` | 2026-06-02 | High | **Full Changelog**: https://github.com/sdsrss/claude-mem-lite/compare/v2.88.0...v2.89.0 |
| `v2.84.2` | 2026-05-25 | High | **Full Changelog**: https://github.com/sdsrss/claude-mem-lite/compare/v2.84.1...v2.84.2 |
| `v2.73.2` | 2026-05-18 | High | **Fix**: regenerated `package-lock.json` via pinned `npm@10.9.2` to restore `@emnapi/core@1.10.0` + `@emnapi/runtime@1.10.0` top-level entries. CI on v2.73.1 went red at `npm ci` because npm@11's `--package-lock-only` strips top-level transitive deps of platform-optional bindings (`@oxc-parser/binding-wasm32-wasi`, `@rolldown/binding-wasm32-wasi`), and GitHub Actions' bundled npm@10 refuses `npm ci` without them. Same recipe documented in `install.mjs:1704` (previously bit #8271 / 2.58.2 / 2.62. |
| `v2.73.0` | 2026-05-13 | High | # v2.73.0 — 9-round dogfood release  End-to-end real-user dogfood pass surfaced 14 distinct issues across CLI, MCP, install, hook, and concurrency surfaces. Mix of **behavior-tightening fixes** (silent fall-throughs that hid typos) and **additive features** (`--json` / `--dry-run` / `--type` on commands users naturally extrapolated from siblings).  Net `+583/-78` across 12 files; `+20` tests (2307 → 2327). Vitest + ESLint green.  ## Highlights  ### Data-integrity / crash fixes  - `update --title |
| `v2.59.0` | 2026-05-08 | High | **Full Changelog**: https://github.com/sdsrss/claude-mem-lite/compare/v2.58.2...v2.59.0 |
| `v2.55.0` | 2026-04-29 | High | **Full Changelog**: https://github.com/sdsrss/claude-mem-lite/compare/v2.54.0...v2.55.0 |
| `v2.53.0` | 2026-04-28 | High | **CLI↔MCP search parity terminator + CLI surface polish.** A QA pass found CLI `search` and MCP `mem_search` returned different ID sets and rankings for the same query — not just presentation differences but actual candidate-set divergence. Root cause: CLI fetched `perSourceLimit = limit` per source while MCP fetched `max(limit*3, offset+limit+10)`, so cross-source merge in CLI mode squeezed observations out of the final top-N when sessions were dense. Plus the two paths had separate inline impl |
| `v2.43.0` | 2026-04-23 | High | **Full Changelog**: https://github.com/sdsrss/claude-mem-lite/compare/v2.42.0...v2.43.0 |
| `v2.34.4` | 2026-04-16 | High | ## Fixed  CJK-short-prompt recall. The T3 raw-length gate (`PROMPT_MIN_LENGTH=15`) in `scripts/user-prompt-search.js` was a raw-character count while the upstream `shouldSkip` already weighted CJK at 3× Latin. A 14-char prompt like `优化 hook 性能降低延迟` (8 CJK + 4 Latin + 2 spaces) passed `shouldSkip`'s 8-unit floor (effectiveLen 30) but fell below the raw-15 gate and never reached FTS.  **No breaking changes.** Latin-only prompts keep identical behavior (`computeEffectiveLen` on pure-Latin text equa |
| `v2.34.3` | 2026-04-16 | High | UserPromptSubmit FTS recall now drops tangential-keyword noise hits via a top-\|rel\| sanity gate. 1670 → 1673 tests green (+3 gate tests, +0.18%).  ## Fixed  - **`scripts/user-prompt-search.js` top-\|rel\| sanity gate added (`TOP_REL_FLOOR=50`).** Noise prompts like `today's date please help me`, `what is the current time right now`, `can you please confirm this works okay` surfaced 3–5 tangential memories each via OR-fallback single-stem matches. Empirical 12-prompt probe: SIGNAL top-\|rel\| 60..133 |

## Dependency audit

- **Score**: 100/100
- **Total deps**: 9
- **Resolved**: 9
- **Unresolved**: 0
- **License conflicts**: 0
- **Warnings**: 0
- **Scanned**: 2026-06-01

## Citation

- HTML: https://www.freshcrate.ai/projects/claude-mem-lite
- Markdown: https://www.freshcrate.ai/projects/claude-mem-lite.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/claude-mem-lite/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._