Concepts

Lore gives AI agents one shared long-term memory system.

It replaces the pattern of copying rules into every local CLAUDE.md, AGENTS.md, and plugin file. Instead, agents read and write a single memory graph that survives sessions, runtimes, and machines.

The short version

• Boot memory provides stable startup context such as shared agent rules, user preferences, and runtime-specific constraints.
• Recall provides task-specific candidate memories for the current prompt.
• Agents should open recalled nodes before treating them as facts.
• Durable changes should be written back into Lore, not copied into every local agent file.
• Dream helps review and improve memory quality, but important changes still need human review.

How agents use Lore

Plaintext

Startup  →  Load boot context (core rules, preferences, channel rules)
Prompt   →  Receive recall candidates for the current task
Work     →  Open relevant nodes, use them as context
Write    →  Save durable facts, decisions, and rules back to Lore
Review   →  Dream checks memory quality, suggests improvements

What to read next

Start with the three concept pages if you are new to Lore:

1. What Lore is for
2. How Lore works
3. When to use it

Then read the memory model pages when you are ready to write better memories: paths, disclosure triggers, priority, and glossary terms.