Why is a chat missing from AI memory?
Check account, capture scope, sync, and retrieval separately when an expected conversation does not appear in AI memory.
Find and fix a missing piece 2 min read
Search with a detail that was unusual in that conversation: a quoted phrase, a project name, an error, or a rejected option. If that fails, change one clue at a time and check whether the conversation was captured at all.
“Marketing” may match half your history. “The pilot starts with three studios” is a better clue if you remember saying it. A file name, an exact error message, or the name of an option you rejected can be more useful than the final topic.
Start with a phrase you remember, then try a broader description of the decision. If you are unsure of the wording, use the purpose: “where we decided not to offer a free trial.” Search results are candidates to inspect, not proof that you found the right decision.
If the source was outside the history range you chose, searching harder will not bring it into the archive. Check the provider, account, capture scope, and sync state. You may need to find the original conversation in the provider first.
Once you find a likely match, read nearby messages and any later correction. Save a short reference in your project brief so you do not have to repeat the same search next week.
TAKE THIS WITH YOU
An illustrative template. Replace the details with material you are allowed to share.
1. Exact detail: “three studios” 2. Project plus topic: pilot audience 3. Decision: why we chose studios for the pilot 4. Rejected option: why we ruled out agencies For each match, check the original conversation and whether the decision changed later.
USING THIS WITH DRAY
Dray can search context from captured work. Coverage depends on the providers and history scope you allowed, as well as successful capture and sync.
Dray is in beta. Available sources and connections depend on your setup. Read the data notes before granting access.
Product details are grounded in the guides below. Example projects and prompts are illustrative, not customer results.