improvement · October 27, 2026
Clearer search, and saves that go through
Search now leads with the better-matching chat, trims extra memories to an excerpt, accepts messy save input, and no longer waits on the memory lookup.
Why we built it
You ask Claude, Cursor, or ChatGPT for last week's thread. Palim used to open with full memories that only shared a word or two, while the chat with that exact title sat underneath. A save could also fail because the assistant sent a number as an id or tags as a comma list. The save failed without telling you why.
Smarter search results
When a saved chat covers your query better than any memory, that chat comes first. Memories that only share a word or two move down and arrive as a 200-character excerpt plus their id. Palim tells you they were cut short because the chats fit better. Search by memory id still returns the whole entry. This is keyword search, not meaning search.
More reliable saving
If Claude, Cursor, or ChatGPT sends a session id as a number, or tags as a comma-separated list, Palim stores the save instead of rejecting it. When the input is still the wrong kind of value, you get a clear message that the field must be text.
Search doesn't wait on memories
The memory lookup no longer holds up the rest of the search.
How to use it
Search and save the way you already do. Ranking, excerpts, and the extra save cases happen on their own. To read a trimmed memory in full, search again with the memory id Palim showed you.
Availability
Live on the hosted MCP at https://api.usepalim.com/mcp, in every connected client.