How Lore works

Understand the loop: boot context, recall, reading memories, writing durable changes, and review.

Lore has a small operating loop.

Operating loop

At startup, the agent receives boot context from fixed memory paths.
For each task, recall searches for relevant memory candidates.
The agent opens the useful memory nodes before relying on them.
After the task, durable changes are written back with path, priority, disclosure, and glossary terms.
Dream and maintenance workflows later review stale, duplicate, or low-quality memory.

The boot baseline is deterministic. Integrations load global nodes such as:

Plaintext

core://agent
core://soul
preferences://user

They also load channel-specific nodes such as:

Plaintext

core://agent/codex
core://agent/claudecode
core://agent/openclaw

Recall is not deterministic startup context. It is a search result for the current prompt. Treat a recall line as a lead, not as proof. Open the node and read the current content before acting on it.

What Lore is forLearn which facts, preferences, decisions, and rules should become durable memory.When to use itDecide which information is worth saving and which short-lived notes can stay out of memory.Memory modelLearn the core memory shape: paths, domains, priority, disclosure, glossary terms, and generated views.Memory pathsUse stable memory paths and domains to keep memory easy to browse, link, and move.