🚀 cognee Update: September 2025

Vasilije MarkovicCo-Founder / CEO

Hey there, cognee community!

September was a sprint. We ran a full Launch Month, upgraded retrieval and graph intelligence, merged a ton of community PRs, hit Github trending again, and showed up at events across the memory/context for AI agents ecosystem. Here’s the quick list and then the details.

At a Glance

  • Launch Month with daily drops: We started with our new evals, product announcements like memify pipeline, auto-optimization, three case studies - still counting.
  • Cognee UI: local & cloud (Cognee Cloud) notebooks plus a Graph Explorer.
  • Retrieval upgrades: a new lexical-chunk retriever, temporal fixes, feedback loops.
  • Graph intelligence: ontology resolver + matching strategies, time-graph options, more.
  • Adapters: new DuckDB vector adapter, Pinecone in community adapters, FalkorDB/Chroma updates.
  • Community wins: we hit GitHub Trending, again!**.
  • We were on the road (and air): talks, conferences, meetups, and partner sessions with Redis, Neo4j, and more.
  • Challenge: the September “Contribute to Win” challenge wraps today—winners announced in the next days.

Launch Month recap

Memify: Memory enrichment layer

Memify lets you treat post-processing as a first-class pipeline: adjust structure, fix entities, or add relationships and persist the deltas—no full rebuild. It’s live with docs + beta access.

Memory Auto-Optimization with Feedback Loops

With cognee's memory auto-optimization, rated responses aggregate into weights on the edges used to answer questions. Those weights guide future ranking so relevance improves with real usage—less manual prompt twiddling.

The new Cognee UI

The new Cognee UI gives you one workspace for adding data, cognifying, querying, and exploring the reasoning subgraph using notebooks. Run locally, hook your cloud datasets, and visualize results in Graph Explorer (enable use_combined_context=True).

Graph-aware embeddings

Introduced Graph-aware embeddings fusing semantic vectors with graph signals (hierarchy, time, entity types) to improve ranking precision, context fidelity, and latency via smarter reranking. Available in paid plans; fully compatible with our open-source stack.

Case studies we shipped

  • Knowunity (EdTech): we mapped 40,000 German learners to surface likely classmates and study partners using proximity + academic context—turning isolated effort into networks. Read more here
  • UWYO (Education policy): turned scattered PDFs into an evidence graph that answers natural-language questions with page-level provenance; built for teacher workflows. Read more here
  • Tier-1 US bank (Financial services): delivered a semantic layer that unifies card rules, APRs, rewards, and fees for precise, auditable answers. Read more here.

Product Highlights

Retrieval & Reasoning & Graph Intelligence

  • Feedback loops enabled during search, scores are written back to enrich the graph
  • Ontology resolver improvements
  • Time graphs: toggle validity windows
  • Temporal retriever improvements: more consistent “as-of” answers.
  • Graph-aware embeddings fuse semantic vectors with graph signals (hierarchy, time, entity types) to improve ranking precision, context fidelity, and latency via smarter reranking.

Cognee UI & DX

  • Programmatic UI start + auto-start backend for frictionless local runs.
  • Simplified search endpoint, graph-view + search harmony.
  • New starter notebook on UI
  • Clearer errors, quieter logs, and CI reliability tweaks.

Adapters & Connectors

  • DuckDB vector adapter
  • Pinecone adapter added to community set.
  • Chroma and FalkorDB updates; adapter tests tightened.

Agent Frameworks / MCP

  • Easier CLI flows to start MCP from the UI, version bumps, and better serialization/return types.
  • We integrated with LangGraph. Many more integrations are in the queue.

Community & Events

  • Our September “Contribute to Win” challenge closes today. We’ll highlight a selection of PRs and announce winners in the next days.
  • We showed up on GitHub Trending—thank you for the stars, forks, and shout-outs from our community and partners!
  • We were out in the wild: Conferences meetups, partner webinars, community newsletters (Year of the Graph), many blog/social media posts from our community.

Some of the highlights:

☑️ AI Alliance × Cognee

☑️ Redis Released – SF

☑️ Graph Exchange (Fall 2025, SF)

☑️ DuckDB Podcast

☑️ AI Engineer Paris

☑️ Qdrant Vector Space Day

☑️ Memgraph community webinar

If you met us at any of these, thank you—these already helped shape our approach this month!

What’s Next

We continue shipping!

Expect many more integrations with Agent frameworks and exciting product announcements.

This month we'll focus even more on building with you, improving with your feedback, refining our docs.

Huge Thanks!

Huge thanks to everyone who shipped PRs, reported issues, and cheered us on at events. You’re steering the roadmap.

Onward!

Cognee is the fastest way to start building reliable Al agent memory.

Try Cognee Cloud Talk to Us

Looking for a custom deployment? Chat with our engineers!Contact us