Mem0 Python SDK v2.2.1 (2026-09-25, PyPI mem0ai 2.2.1) makes `Memory`/`AsyncMemory.add()` report only vector-store inserts that actually succeeded—raising `VectorStoreError` when none insert—restores context-manager close semantics, fixes Bedrock Anthropic Converse to read the first text-bearing content block (after optional reasoningContent), applies full Turbopuffer filter operators, and remaps euclidean(_squared)/S3 Vectors scores to `1/(1+distance)`. Same-day Node SDK ts-v3.3.1 stops injecting OpenAI default baseURL/model into non-OpenAI providers.

Key Takeaways

  • ✓Shipped: Python v2.2.1 / PyPI mem0ai 2.2.1; Node ts-v3.3.1 same day
  • ✓Correctness: add() returns only inserted memories; raises VectorStoreError if none insert
  • ✓Bedrock Anthropic: read first text-bearing Converse block after optional reasoningContent
  • ✓Turbopuffer filters: full eq/ne/gt/gte/lt/lte/in/nin; unsupported ops raise ValueError
  • ✓Scores: euclidean_squared and S3 Vectors euclidean map via 1/(1+distance) to avoid negatives/collapse-to-zero
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
🐙
GitHub Release Mem0 Python v2.2.1
github.com/mem0ai/mem0/releases/tag/v2.2.1
↗
📖
PyPI mem0ai 2.2.1
pypi.org/project/mem0ai/2.2.1/
↗
🐙
GitHub Release Mem0 Node ts-v3.3.1
github.com/mem0ai/mem0/releases/tag/ts-v3.3.1
↗
📖
Mem0 Docs
docs.mem0.ai/
↗
🐙
mem0ai/mem0 repository
github.com/mem0ai/mem0
↗
🐙
v2.2.0
github.com/mem0ai/mem0/releases/tag/v2.2.0
↗
🔬

In-Depth Technical Analysis

Core Background & Industry Pain Points

Agent long-term memory needs truthful writes and ranked retrieval. Mem0’s add() could treat vector-store rejections as successful ADDs (history/entity links/returns), Bedrock Claude reasoning models emit reasoningContent before text so content[0] KeyError’d, Turbopuffer silently dropped non-gte/lte filters, and 1-distance / S3 Vectors euclidean scoring produced negatives or mass zeros.

Architecture Highlights & Internals

v2.2.1 (2026-09-25, PyPI mem0ai 2.2.1) returns only inserted memories and raises VectorStoreError when none insert; restores context-manager close; Bedrock Anthropic reads the first text-bearing Converse block; Turbopuffer supports eq/ne/gt/gte/lt/lte/in/nin; euclidean_squared and S3 Vectors euclidean scores use 1/(1+distance). Same-day ts-v3.3.1 stops injecting OpenAI default baseURL/model into non-OpenAI providers.

Authoritative Benchmarks & Measured Scores

No public Recall/latency numbers—this is a correctness release versus v2.2.0. Validate by fake-ADD disappearance, Bedrock reasoning path success, and filter/score ranking regressions—not throughput claims.

Developer Hands-on Guide

pip install -U mem0ai==2.2.1, catch VectorStoreError around add(), retest Bedrock Claude reasoning Converse, and verify Turbopuffer eq/in plus euclidean_squared ranking. Node: npm i [email protected]. Docs: docs.mem0.ai; release: v2.2.1.