Memory model
Learn the core memory shape: paths, domains, priority, disclosure, glossary terms, and generated views.
The memory model is path-native. Every memory lives at a stable URI and carries the metadata needed for recall and review.
Memory fields
Each useful memory should have:
| Field | Purpose |
|---|---|
| URI/path | stable address such as project://my_app/deployment |
| content | the durable fact, rule, decision, or workflow |
| disclosure | when this memory should appear in the future |
| priority | how important it is relative to other memories |
| glossary | aliases and terms users or agents are likely to type |
| history | change record for review and rollback |
Paths make memory browsable. Disclosure and glossary make it recallable. Priority helps ranking but cannot substitute for good content.
Domains
| Domain | Use for |
|---|---|
core:// | agent-wide rules, runtime-specific rules, and shared operating context |
preferences:// | stable user preferences and durable personal context |
project:// | project-specific facts, decisions, commands, constraints, and workflows |
Keep memory focused. One memory should describe one durable topic. If a note covers several projects, runtimes, or decisions, split it.