cognee

The cognee team will be at In The Loop by n8n in Berlin on Tuesday and Wednesday. Come say hi.

We’re partnering with Briefcase and Stripe on this amazing hackathon in London next Thursday. Theme is: AI agents in fintech. Three hours to build an agent that handles real financial operations. Judges are: • Ján Stehlík (Briefcase) • Fionn O Colmain (Stripe) • David Gelberg (10 Downing Street AI Team) • Dominic Hudson (AI at Meta) • Akash Bajwa (Earlybird Venture Capital) Thursday 15 October, 17:15 to 22:00 BST, Central London. Win a price of £2,000. Free, apply here: https://luma.com/3k1pazop

Long context is valuable. It helps when the relevant information can be passed directly into the model. But context still has limits. It disappears unless something stores it. Long prompts increase cost and latency. Models do not use long context perfectly. Long-running tasks eventually produce more information than one request should carry. Memory handles a different job: deciding what should persist between calls, sessions, agents, and tools. The context window is just the delivery mechanism.

Two systems can get the same final-answer score while maintaining very different memory stores. One may have clean entities, scoped facts, provenance, and update rules. Another may have duplicate customers, unresolved contradictions, stale preferences, and broken links that happen not to show up in today's eval. Production memory needs evaluation at multiple levels. Did the agent answer or act correctly? Is the memory state itself healthy enough to keep using tomorrow?

A memory benchmark is most useful when the system cannot simply pass the whole history into the model. That is why benchmark scale matters. LoCoMo is useful for conversational recall, but the public conversations average around 26k tokens. BEAM is more demanding for memory architecture because it scales coherent histories up to 10M tokens. But benchmark scores are still directional. For production systems, the final question is whether memory improves the real task: completion rate, fewer turns, better personalization, fewer repeated mistakes, lower support burden, or better conversion.

What agent memory actually means An LLM does not automatically remember previous work. Each model call is stateless: the model receives a prompt, generates an answer, and the next call starts independently. Continuity comes from the system around the model. That can include a session buffer, context engineering, RAG, and a long-term memory layer. The difference with agent memory is that the system can accumulate new information over time, retrieve it when relevant, update it when facts change, and forget it when it should no longer be used. Memory is not more tokens. It is managed state.

Employees spend years learning how knowledge moves through a company: what matters, where it lives, who needs it, and what they’re allowed to access. AI agents face the same challenge. As companies adopt more agents across engineering, product, sales, and support, the real challenge isn’t giving every team an agent. It’s making sure those agents work from the same company context. Our CEO, Vasilije Markovic, shows how cognee turns fragmented company knowledge into shared memory that agents can actually build on.

Keenable is now a web-fetch backend in cognee Name what you need in the fetch call, and only that part of the page is chunked, embedded and left in the knowledge graph.

An agent that reads a web page is reading navigation, cookie banners, scripts and footer links too. And whatever it pulls out of that is gone at the end of the session. Keenable.ai is now a built-in web-fetch backend in cognee, from version 1.5 on. Keenable.ai takes a URL and returns the page as clean Markdown. cognee takes that Markdown into a persistent knowledge graph, so the agent can still query what it read weeks later. Setup is one environment variable: KEENABLE_API_KEY="keen_..." Then a URL goes straight into the memory workflow: await cognee.remember(url) await cognee.recall("what does the enterprise tier cost?") There is no Keenable SDK to install. cognee uses the HTTP client it already ships with. For pages that keep changing, the scheduled scraper runs the same fetch path on a cron and hashes every page, so a run can tell whether the content moved. How it works end to end: https://lnkd.in/dbznGeWg

Make sure to have a look at this post: