Thenwise Adds Cross-Session Memory to Claude Code: 83% vs 47% Accuracy, 40% Fewer Tokens
dheeraj_ponnaganti · reddit · 2026-10-30
A developer released Thenwise, a free open-source Claude Code plugin that gives Claude local memory of past chats, commits and notes, so it can answer questions like "why did we split the cache?" with dated sources.
How it was built (with Claude Code itself):
- Claude wrote most of the TypeScript: local store, search, an MCP server with seven memory tools, and hooks that save each chat
- A separate Claude session reviewed each release for bugs, catching race conditions and sync edge cases
- Eval: run Claude Code on hidden-answer questions about real project history with and without Thenwise, blind-graded
- Dogfooding: Thenwise was installed while building Thenwise, surfacing rough edges fast
Results: 83% accuracy vs 47% for plain Claude Code, 40% fewer tokens, 2.2x faster answers.
Apache 2.0, no paid tier, no telemetry; API keys are masked and [no-memory] opts a chat out of storage.
More from coding & agent
- pi-durable performance pass: every agent step saved as a commit for crash resume — mitsuhiko · 2026-10-30
- AI Agent Explained in One Video: Brain, Tools, and a Loop — faraz0x · 2026-10-30
- PCRE2 reimplementation in Verus Rust has bugs you can't file upstream issues for — jd_pressman · 2026-10-30
- Apple Trains UI Navigation Agents with Online RL Directly on Live iOS Apps — apple · 2026-10-30
- Cursor confirms trojanized Claude Code v2.1.313 extension hit up to 5,000 users with SSH backdoor — cyb3rops · 2026-10-30
- Cartesia launches Managed Agents, tops Speech Agent Arena for enterprise voice AI — anthara_ai · 2026-10-30