PRs to come from me on openmemory - Roadmap

[frederikb96]

Hi @parshvadaftari,

I wanted to give a small outline on the other PRs that are still to come from me over the next time as the merging progresses 🙂

Here a quick and dirty screenshot as an overview... All the commits tagged with PR are the ones I plan to create PRs for soon. See below.

Here are some smaller ones (there are some more though; I didn't write all down here yet):

Critical Bug Fixes:

1. Fixing config trailing slash issue
2. Fixing the memory add method which is currently broken since it only stores metadata in Qdrant but not in the OpenMemory database
3. Fixing the REST update_memory API

Non-Breaking Enhancements:
4. Add limit parameter to MCP search

5. Making the whole OpenMemory part async instead of sync. Because I had huge issues searching memories and using the GUI since it's blocking during memory creation. The problem is that the API layer (MCP + REST) isn't awaiting the async memory_client calls, so it blocks the event loop causing the UI to freeze. NO breaking change though - just a lot of small changes in code everywhere.

6. Non-breaking feature/fix on date filtering in Qdrant. I implemented an opt-in method to use real numeric date filtering on the Qdrant layer. I needed this desperately to be able to have real date filtering in search. The background: ISO string timestamps in Qdrant without payload schema means lexicographical comparison which is broken for dates. So I added numeric timestamps (created_at_ts/updated_at_ts) like Redis/Valkey does. Only drawback - all memories ingested before that feature don't have the numeric timestamp and I was too lazy implementing a migration. So it's a non-breaking change with optional opt-in to use the new numeric timestamp filter method in Qdrant. It's super well abstracted from all the rest of the code. Minimally invasive and doesn't break anything. All new data ingested since it's implemented automatically supports both filter modes, new and old one. Even if one decides to opt in to new filter mode at a later point in time.

Breaking Changes:
7. BREAKING: Complete rewrite of the OpenMemory MCP server - from deprecated SSE to Streamable HTTP. This fixes security vulnerabilities, updates the outdated MCP library to current spec, removes multiple lines of legacy code. Huge improvements for usability and brings it up to date with current MCP server standard. This is the ONE large breaking change I have made. However, I guess there is no other way going forward, and the user base of the OpenMemory subproject is probably not large yet. So I hope that you're fine with merging this.

Moreover, all my other improvementes and fixes are based on this new MCP server implementation 😅