Ai

Mem0 v2.0.3: Enhanced Filter Operators for PGVector and Server Fix

Mem0 v2.0.3: Enhanced Filter Operators for PGVector and Server Fix

Mem0 just dropped v2.0.3 of its Python SDK, and it’s a meaningful update for anyone building AI applications on top of vector databases. The headline? PGVector now supports rich filter operators — things like gt, lt, wildcards, and logical $or. Previously, if you passed an operator dict to a search or list call, it would get silently stringified and return zero results. That bug is gone. There’s also a fix for the /search endpoint that was returning 502 errors under some server configurations. Let’s break down what changed and why it matters.

What Changed

The PGVector adapter now supports a full suite of filter operators in search(), keyword_search(), and list(). You can use eq, ne, gt, gte, lt, lte, in, nin, contains, icontains, wildcard *, $or, and $not. Before v2.0.3, only exact-equality filters worked — operator dicts were ignored (stringified into the query but not executed), leading to zero results and a lot of head-scratching. The fix is in pull request #5263.

Separately, the server’s /search endpoint was returning HTTP 502 errors when something went wrong internally. That's now resolved, though the release notes don't specify the root cause. It’s a stability patch, pure and simple.

Why It Matters

For AI engineers using Mem0 for memory or retrieval-augmented generation, this is a big deal. Rich filters mean you can now do precise vector searches — for example, find all memories where the timestamp is greater than a certain date, or where metadata contains a keyword but excludes another. That opens up more complex query logic without needing to pull all results and filter client-side.

I’ve been watching Mem0 evolve, and this fix feels overdue. The silent stringification bug was the kind of thing that would waste hours of debugging. It’s good to see the team prioritizing developer experience. The server fix, while less glamorous, matters for production deployments — nobody wants a 502 ruining an inference pipeline.

Overall, v2.0.3 is a small but focused release. If you're using PGVector with Mem0, upgrade today. Your queries will finally work as expected.

Official Source: https://github.com/mem0ai/mem0/releases/tag/v2.0.3