Frequently asked questions about memory
How long are memories stored?
How long are memories stored?
Generated fragments and profiles have no expiry date (unless you configured an expiry when creating the memory). You can delete unneeded memories anytime via console or API.
How is memory isolated between users?
How is memory isolated between users?
Memory storage and retrieval are isolated by entity (
user_id). Different users' memories don't interfere. You can also create multiple memories for different business scenarios.What's the difference from the previous long-term memory?
What's the difference from the previous long-term memory?
Memory is a full upgrade of the previous long-term memory. Key differences: memory is an independent management container, supports more flexible rule configuration, adds user profiles, and optimizes retrieval quality, cost, and latency.
Fragments vs. profiles — when to use which?
Fragments vs. profiles — when to use which?
Fragments suit specific events and info (e.g., "User went to Beijing last week"). Profiles suit structured attributes (age, occupation, preferences). Use profiles for fixed attributes, fragments for dynamic events. Both can be used together.
Where can I see memory entity counts?
Where can I see memory entity counts?
On the memory list page, click View details. The Memory details tab shows counts for fragments, profiles, and entities.
How do I optimize retrieval?
How do I optimize retrieval?
Test in the Memory search tab; adjust the similarity threshold (recommended 0.5~0.7); enable intent recall to avoid irrelevant searches; enable rewrite and rerank for accuracy; set max results appropriately.
Can the default memory be deleted?
Can the default memory be deleted?
The default memory cannot be deleted, but you can edit its name, description, and rules. It has a preset "Default project" fragment rule with 180-day expiry that you can modify or supplement.
Are there API rate limits?
Are there API rate limits?
All endpoints share a 3000 QPM limit (account level). Fragment add is 120 QPM, fragment search is 300 QPM. See Rate limits.
What's the difference between Pro and Lite? How to choose?
What's the difference between Pro and Lite? How to choose?
Pro enables Rerank (result reranking) for higher quality but higher cost. Lite disables Rerank for lower cost. Use Pro for accuracy-critical scenarios, Lite for high-frequency, cost-sensitive ones. Add call version is set by the rule's
plan_version. Search call version is controlled by the request parameter plan_version. See Billing.How long is the free quota valid?
How long is the free quota valid?
Add and Search free quotas are valid for 3 months from the commercialization date. 10,000 free storage entries have no expiry.
Can I switch between Pro and Lite after creating a rule?
Can I switch between Pro and Lite after creating a rule?
Yes. Edit the rule in the console to change the plan version, or use the UpdateProfileSchema API to modify the
plan_version field. New writes follow the updated version; existing memories are unaffected.