OpenLTM brings local, self-decaying memory to AI coding agents
OpenLTM is a newly open-sourced, MIT-licensed long-term memory plugin for AI coding agents like Claude Code, OpenCode, and Pi, storing persistent semantic memory in a local SQLite database with no cloud or telemetry.
Score breakdown
OpenLTM addresses a core limitation of AI coding agents — the loss of project context across sessions — by providing a fully local, open-source memory layer with importance-weighted decay and semantic recall.
- 01OpenLTM is a long-term memory plugin supporting Claude Code, OpenCode, and Pi.
- 02It was originally a private plugin and has now been fully open-sourced under the MIT license.
- 03Memory is stored locally in a SQLite database — no cloud, no account, no telemetry.
OpenLTM started as a private memory layer for Claude Code, built to solve the problem of AI coding agents repeatedly asking developers to re-explain their codebase across sessions. It has now been released as fully open source under the MIT license, with the repository moved to `RohiRIK/OpenLtm` and the plugin renamed to `openltm`. Existing memory databases from earlier installs carry over automatically.
The system provides persistent semantic memory — including automatic capture, semantic recall, importance-weighted decay, and a queryable memory graph — that survives every session, update, and compaction.
The system provides persistent semantic memory — including automatic capture, semantic recall, importance-weighted decay, and a queryable memory graph — that survives every session, update, and compaction. All memory is stored in a local SQLite database owned by the user, with no cloud backend, no account requirement, and no telemetry. The architecture is provider-agnostic, built around a single core package (`@rohirik/openltm-core`) with thin adapters for each supported host. Hooks, skills, janitor providers, and a graph visualizer are all included in the open repository.
Key facts
- 01OpenLTM is a long-term memory plugin supporting Claude Code, OpenCode, and Pi.
- 02It was originally a private plugin and has now been fully open-sourced under the MIT license.
- 03Memory is stored locally in a SQLite database — no cloud, no account, no telemetry.
- 04Features include automatic capture, semantic recall, importance-weighted decay, and a queryable memory graph.
- 05Architecture uses a provider-agnostic core (`@rohirik/openltm-core`) with thin adapters per host.
- 06The repository has moved to `RohiRIK/OpenLtm` and the plugin is now named `openltm`; existing memory databases carry over.
- 07The project includes hooks, skills, janitor providers, and a graph visualizer.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 9, 2026 · 17:05 UTC. How this works →