improve curator interface
S
Steve Berry
In trying to use chat to mass edit and adjust tags, the chat frequently claimed a 'curator' failed with frustratingly small requests. Complaining to max out around 100 cards and struggling to match with basic card title prefixes. At one point completely deleting a tag instead of just removing it from instructed cards.
This is not sufficient performance from a service centered around knowledge base work.
Log In
Sankari Nair
Hey Steve,
I'm so sorry about the issue you're having. Could you please email us at support@recall.it with a bit more detail? We can dig into how we can support you here.
Please note that on our Max plan, you can bulk select and tag 100 cards at the same time through AI generation. You're right that it is limited to 100, and the main reason is the AI costs associated with this action.
S
Steve Berry
The issues with the 100 card limit I can understand, i just didn't catch that the tagging operation would fall under that category. However, the issues for this particular conversation involved card sets that were <30. Summery below:
### 1. Curator Failure with Multi-Step Instructions
The background curation system repeatedly failed to generate a plan when asked to perform a combined action (e.g., "Create a new tag AND assign it to specific cards"). Even when provided with exact card titles and a clear scope, the system returned a generic "could not produce a plan" error.
*
Workaround required:
The task had to be split into two separate turns—first creating the tag, then assigning it.### 2. Tag Name Sensitivity
The curator failed to process a tag creation request for a long, descriptive name (
Source: Ochoa and Osier - The Writer's Guide to Creating a Science Fiction Universe
). *
Workaround required:
The name had to be significantly shortened (Source: Ochoa and Osier
) before the system would accept the instruction.### 3. Over-Destructive "Delete" Logic
When instructed to "remove a tag from all cards," the curator interpreted this as a command to delete the tag itself from the knowledge base.
*
Consequence:
This caused a "cascading failure" where the parent tag was deleted, and all its subtags were orphaned and moved to the top level of the tag tree, requiring manual reorganization to fix.### 4. Inconsistent Validation
The system occasionally reported that a goal was "too broad" despite being scoped to a specific tag and a specific set of cards with matching strings in their titles. This suggests a lack of precision in how the curation model parses bounded contexts.
---
Now, this could simply be that I overestimated the capabilities of what is available to us on the Plus plan, but I would be hard pressed to put these operations in the 'bulk data operation' category for Max plan usage.
Also, for clarification, does the plus plan even have access to any models above say the Claude Sonnet level? I have a hard time distinguishing whether the Max plan allows us to SPECIFY particular frontier models or if it is required to have access to them AT ALL. If that makes sense.