What is Lore
Lore gives your AI agents one shared long-term memory: a stable baseline at session start and the right memories before every reply.
Rules copied across CLAUDE.md, AGENTS.md, and plugin prompts drift out of sync, and each copy goes stale on its own. What one agent learns stays locked in its own chat, so the next agent has to learn it again.
What Lore does
Every connected agent runs the same loop:
- Boot. At session start, Lore sends a fixed baseline: shared workflow rules, the persona, your stable preferences, and the rules for the current runtime.
- Recall. Before each prompt, Lore adds a short
<recall>block with the memories that might matter. - Read before trust. A recall line is a lead, not a fact. The agent opens the node with
lore_get_nodebefore it relies on the content. - Write back. When the agent learns a durable fact, it saves the fact as a memory node with the memory tools.
- Dream. On a schedule (03:00 by default), Dream reviews the memory graph, fixes duplicates and vague triggers, and records every change.
Per-node history and backups make these changes reversible.
Key ideas
| Concept | What it means | Read more |
|---|---|---|
| Memory node and path | A plain-text memory with a stable address, such as project://checkout_api/webhooks. | Memory model |
| Boot | The fixed baseline Lore sends at session start. | How Lore works |
| Recall | Candidate memories that match the current prompt. | How recall works |
| Disclosure and glossary | When a memory should surface, and the words that point to it. | Disclosure and glossary |
| Priority | An integer rank: 0 for core identity, 1 for key facts, 2 and up for general memories. | Priority |
| Dream | Scheduled maintenance that keeps the memory graph clean. | Dream |
| Skills (Cloud) | Reusable instruction packages that agents load when they are relevant. | Skills |
Two ways to run Lore
Both options use the same agent plugins. The installer connects your agents to whichever server you choose.
| Loremem Cloud | Self-hosted | |
|---|---|---|
| Who runs it | Loremem, at https://app.loremem.com | You, on your own machine or server |
| Setup | Sign in, create a token, and run the prefilled install command | Run npx @loremem/cli, choose Docker, then finish setup at /setup |
| Models | Preconfigured. Choose an embedding profile and a runtime LLM profile in Settings → Models. No API keys needed | Bring your own OpenAI-compatible embedding endpoint and a chat model for View LLM |
| Skills | Yes | No |
| Where data lives | Loremem's hosted service | Your machine or server |
| Cost | Free plan. Sign in with GitHub or Google | Free and open source (MIT). You pay for your host and any model APIs you use |
Note
You can move a self-hosted memory to Loremem Cloud later by importing a backup. See Import a self-hosted backup.
Where to go next
- Quickstart: connect your first agent and save a memory.
- How Lore works: what Lore sends and records during a session.
- Connect agents: setup for each runtime and for other MCP clients.
- Writing good memories: the shape of a memory that recall can find.