OpenLTM – Local, self-decaying memory for AI coding agents

Uh oh!

There was an error while loading. Please reload this page.

Notifications You must be signed in to change notification settings

Fork 0

Star 13

BranchesTags

Open more actions menu

Folders and files

NameName

Last commit message

Last commit date

Latest commit

History

243 Commits

243 Commits

.claude-plugin

.claude-plugin

.github

.github

agents

agents

assets

assets

commands

commands

docs

docs

graph-app

graph-app

hooks

hooks

migrations

migrations

packages

packages

scripts

scripts

skills

skills

src

src

.env.schema

.env.schema

.gitignore

.gitignore

.npmrc

.npmrc

.pre-commit-config.yaml

.pre-commit-config.yaml

.trufflehogignore

.trufflehogignore

AGENTS.md

AGENTS.md

CHANGELOG.md

CHANGELOG.md

CLAUDE.md

CLAUDE.md

CONTRIBUTING.md

CONTRIBUTING.md

LICENSE

LICENSE

README.md

README.md

bun.lock

bun.lock

bunfig.toml

bunfig.toml

install.sh

install.sh

package.json

package.json

tsconfig.json

tsconfig.json

Repository files navigation

You explained your auth layer once. Why does Claude ask again tomorrow?

Long-Term Memory for AI coding agents — Claude Code, OpenCode, and Pi

Persistent semantic memory that survives every session, every update, every compaction.

📣 Now open source

OpenLTM was born as a private memory layer for Claude Code. Today it's fully open source under MIT.

Same engine — automatic capture, semantic recall, importance-weighted decay, a queryable memory graph — now yours to read, fork, and extend. What started as one developer's "stop re-explaining my codebase" plugin is now an open foundation for agent memory across Claude Code, OpenCode, and Pi.

🔓 MIT licensed — no cloud, no account, no telemetry. Your memory lives in a local SQLite DB you own.

🧩 Provider-agnostic — one core (@rohirik/openltm-core), thin adapters per host.

🛠 Hackable — hooks, skills, janitor providers, and the graph visualizer are all in the open.

Migrating from an earlier install? The marketplace is now RohiRIK/OpenLtm and the plugin is openltm. Your existing memory database carries over.

The philosophy

Four ideas. No exceptions.

Memory should be automatic. Hooks do the work. The session end hook extracts patterns, the session start hook injects them back. You shouldn't have to remember to remember.

Decay is a feature, not a bug. A gotcha from six months ago that you never revisited probably no longer applies. Set importance: 5 to make something permanent — everything else ages out naturally.

Semantic over keyword. FTS5 full-text search runs first; if it returns nothing, vector embeddings kick in. You search by meaning, not exact words — "how we handle async errors" finds the right memory even if you never wrote those exact words.

Zero config, zero lock-in. Install once, works everywhere. Every setting has a sane default. The DB lives outside the plugin directory so it survives every update. No cloud, no telemetry, no account.

What you get

🔍 Recall Past decisions, patterns, and gotchas — before you start work

🧠 Learn Every session, automatically — no manual note-taking

💉 Inject Top context at session start so Claude picks up where it left off

⏳ Decay Stale memories fade while critical knowledge lives forever

🕸 Graph Traverse relationships between memories for reasoning chains

🗺 Visualize See your entire memory network in a browser-based explorer

⚡ Vec Recall Semantic vector (KNN) recall via sqlite-vec; degrades to JS-cosine when unavailable

🔌 Extensions sqlite-vec + Honker (queue/cron/pub-sub) loaded dynamically; graceful fallback without system libsqlite3

Install

Marketplace (recommended)

claude plugin marketplace add https://github.com/RohiRIK/OpenLtm claude plugin install openltm

Restart Claude Code. That's it. Four Claude Code hooks + one git post-commit hook auto-wire, four commands load, five skills activate, and your openltm.db migrates or creates itself.

bunx (no clone)

bunx @rohirik/openltm-core # auto-detect Claude Code, OpenCode bunx @rohirik/openltm-core --pi # experimental Pi adapter bunx @rohirik/openltm-core --dry-run --claude # preview without writing

Dev / git clone

git clone https://github.com/RohiRIK/OpenLtm ~/Projects/OpenLtm cd ~/Projects/OpenLtm && bash install.sh

Quick Start

Start a new session. Context is injected at the top automatically.

Then try:

/openltm:memory recall auth — what do we know about auth in this project? /openltm:memory learn — save something worth keeping /openltm:health — memory health + decay summary /openltm:project init — set a goal for the current project

That's it. The rest is hooks doing the work.

The shape of memory

Claude Code │ ├── 4 Commands ──┐ ├── 5 Skills ──┼──▶ openltm MCP server ──▶ openltm.db └── 5 Hooks ──┘ (memories, tags, context_items, memory_relations, memories_fts)

Full deep-dive — schema, hook architecture, decay formula, ADRs — in How It Works and Architecture.

SQLite Extensions: The plugin loads sqlite-vec (vec0 / KNN vector search) and Honker (async embedding queue, leader-elected cron, pub-sub) when a system extension-enabled libsqlite3 is available. Both degrade gracefully — missing binary or library leaves the capability off and falls back to JS-cosine, file-watch polling, or in-process cron. Controlled with LTM_DISABLE_VEC, LTM_DISABLE_HONKER, LTM_SQLITE_LIB, and LTM_HONKER_EXT env vars.

Verify

/openltm:health # plugin health + hooks + decay /openltm:memory recall test # returns results (or "no results" on fresh install)

Start a new session — you should see context injected at the top. If not, run /openltm:health to diagnose.

Go deeper

Full documentation index: docs/.

I want to… Read

Get running in five minutes Quickstart

See every install option Installation

Use every command and its flags Commands

Tune decay, injection, embedding behavior Configuration

See how it works under the hood How It Works · Architecture

Understand the schema and data model DB Spec

See all hooks, skills, and MCP tools Hooks · Skills · MCP Tools

Fix a problem Troubleshooting

See the product vision and where it's going PRD · Roadmap

Contribute a change Contributing

Check what changed Changelog

Contributing

Open an issue first to discuss the change, then send a PR. The full workflow — setup, tests, the version-bump rule, and release steps — is in CONTRIBUTING.md.

Report a Bug

License

MIT — RohiRIK

Built for Claude Code

Powered by caffeine, SQLite, and the persistent belief that context shouldn't die at the end of a session.

About

Long-Term Memory plugin for Claude Code — semantic search, context injection, session learning

Resources

Readme

License

MIT license

Contributing

Contributing

Uh oh!

There was an error while loading. Please reload this page.

Activity

Stars

13 stars

Watchers

0 watching

Forks

0 forks

Report repository

Releases 11

v2.9.0

Latest

Jun 8, 2026

+ 10 releases

Sponsor this project

Uh oh!

There was an error while loading. Please reload this page.

Packages 0

Uh oh!

There was an error while loading. Please reload this page.

Contributors

Uh oh!

There was an error while loading. Please reload this page.

Languages

TypeScript 99.3%

Other 0.7%