Yesterday, we released our paper, "Optimizing the Interface Between Knowledge Graphs and LLMs". We have developed a new tool to enable AI memory optimization that considerably improve AI memory accuracy for AI Apps and Agents. Let’s dive into the details of our work 📚

Some days it is like this, except for the fact that nobody is Austrian and the language is English

We just released the Redis adapter for @cognee_ - another great vector database in our poly-store toolbox. What I love: millisecond recalls, hybrid filters, dead-simple RedisVL setup. Install with: pip install cognee-community-vector-adapter-redis. Full walkthrough & code here.

Activity while at cognee this year:

Github activity when working in a regular job

Pinki the office dog puked on this story, read more here →

👀 Want zero-setup? Sign up for SaaS beta waitlist is LIVE—get advantage of being early birds:

Numbers don’t lie: 6k⭐, dozens of PRs, record number of pipeline runs. Thanks to every contributor! 💜

Top highlights 🚀. – Community ↔ Core split → free & hack-friendly and production ready.– RBAC, API keys, audit logs → security by design.– Redis, Azure Vector Search connectors.– MCP 2.0 + Associative Memory = next-level reasoning 🤯.

June drop is here! 🧠⚡️. @cognee_ just shipped a massive update: .- Community vs Core split.- Qdrant partnership.- secure permissions.- Associative Memory that connects concepts on the fly. Blog + beta waitlist👇.

Death Star? . No, just another project for a partner we are working on. Stay tuned, soon in our case studies!

You know you confused Reddit when it starts suggesting .r/azerbaijan in the main feed.

I'd never hire someone who has "coach" in their LN description + "online" version of a prestigious university. All the wrong signals.

Ready to cook? Full tutorial 👉

Bonus: You can cron the whole thing with one line in Towerfile - perfect for daily digests or auto-refreshing agent memory ✨.

Why care?. • One catalog, zero copying.• Sub-second graph search.• Declarative scheduling in Tower.• Graph + tables stay version-controlled.

Key loop (all in Tower):. 1️⃣ Save HackerNews stories → Iceberg.2️⃣ Pull with Polars.3️⃣ Cognee extracts entities & links.4️⃣ Ask natural-language questions.5️⃣ Push answers back to Iceberg ♻️.

🔹 Iceberg = ACID tables + time-travel on cheap object storage. 🔹 Tower = server-less Python jobs that speak Iceberg natively. 🔹 Cognee = “cognifies” data into queryable memory.

The pain: object storage, catalogs, engines, AI tooling… all glued by midnight YAML. We turned that spaghetti into one pipeline that stores → reads → processes → searches → persists—no Spark, no Airflow.

Why should your lakehouse also be your AI memory?. We just shipped a hands-on guide that fuses Iceberg, Tower & cognee so agents can remember at scale. Let’s dive in 🧵👇.