Persistent memory for AI coding agents.
Lore captures skills, conventions, and project knowledge as you work — then loads them every session so your agent starts with full context instead of starting over. Plain files, git-versioned, zero runtime dependencies.
npx create-lore my-project
cd my-projectThen open a session in your editor:
| Platform | Command |
|---|---|
| Claude Code | claude |
| Cursor | Open the project — hooks activate via .cursor/hooks.json |
| OpenCode | opencode |
No configuration needed. Your first session gets a full context banner immediately.
Sessions accelerate instead of resetting. Every session opens with your project identity, conventions, active work, available agents, and a map of everything your agent knows. No re-explaining.
Semantic search and a live docs UI — highly recommended. Run /lore-docker to start a local Docker sidecar that gives agents semantic search over the full knowledge base and opens a MkDocs site at localhost for browsing it visually. Without Docker, agents fall back to Grep/Glob search.
Gotchas become skills that persist. When your agent hits an API quirk, an encoding edge case, or a deployment gotcha, it captures that as a skill. That skill loads in every future session. The mistake happens once, the fix persists.
Conventions are enforced, not just documented. Your coding standards, docs rules, and security policies are injected before every file write. The agent sees the relevant rules right when it matters.
One knowledge base, every platform, every repo. Capture a skill in Claude Code — it's available in Cursor and OpenCode. Link repos to one hub — they all share the same knowledge. No copying, no drift.
Complex work delegates to focused workers. When work benefits from a fresh context window, the orchestrator spawns workers loaded with curated skills and conventions. Compound tasks split across parallel workers for maximum throughput.
Without Lore — Every session starts cold. You re-explain your project, the agent re-discovers API quirks, makes the same mistakes, and yesterday's debugging session is gone.
With Lore — The agent knows your project. Skills from last week load automatically. Conventions are enforced at write-time. Active roadmaps surface at startup. Complex work delegates to focused workers.
Lore is a directory of markdown files, hooks that shape agent behavior, and scripts that keep everything consistent.
- Skills (
.lore/skills/) — Gotchas and patterns captured from real work. Available on-demand in every session. - Agents (
.lore/agents/) — Worker agents for delegated tasks, loaded with relevant skills per-task. - Docs (
docs/) — Project context, conventions, environment knowledge, runbooks, and work tracking. - Hooks (
.lore/hooks/,.cursor/hooks/,.opencode/plugins/) — Inject context at session start, enforce conventions before writes, nudge knowledge capture during work. - Scripts (
.lore/scripts/) — Platform sync, validation, nav building.
All hooks are plain JavaScript you can read in minutes. They don't make network requests, execute shell commands, or access anything outside your project directory.
Link work repos to a central Lore hub so hooks fire from the hub even when you open the work repo directly:
/lore-link ~/projects/my-app
One hub, many repos, shared knowledge. See the cross-repo guide.
| Command | What it does |
|---|---|
/lore-capture |
Review session work, capture skills, update registries, validate consistency |
/lore-consolidate |
Deep health check — stale items, overlaps, knowledge drift |
/lore-link <path> |
Link a work repo to this hub |
/lore-docker |
Start/stop the local Docker sidecar — semantic search + live MkDocs UI |
| Platform | Integration | Status |
|---|---|---|
| Claude Code | Hooks + CLAUDE.md |
Stable |
| Cursor | Hooks + MCP server + .mdc rules |
Stable |
| OpenCode | ESM plugins + opencode.json |
Stable |
All platforms share the same knowledge base. Skills, agents, and conventions written once sync to platform-specific formats automatically.
Full docs: lorehq.github.io/lore-docs
See CONTRIBUTING.md for development setup and guidelines.
Security issues: see SECURITY.md.