Running /attach is very slow for large datasets

[sarahwooders]

Now that postgres is integrated, MemGPT can support large archival memory stores. However, loading data into the agent via /attach requires copying data from one table into another (the agent's archival memory table).

Possible solutions:

• Parallelizing the copy across multiple processes
• Avoiding having to copy data by allowing agents to have read-only access to attached data source tables (this would require supporting the ability for archival memory to search across multiple tables)

[github-actions]

This issue has been automatically closed due to 60 days of inactivity.