Memory & search

Turns documents and chats into a knowledge graph agents can remember from.

The memory layer of the agent stack — where a knowledge graph, not a bigger context window, is the bet on what an agent should remember, and where the open-source project and the enterprise deployment have to stay the same product.

Makes a company’s files searchable by agents, with the page each answer came from.

The retrieval layer an agent actually runs on — they train the embedder, the reranker and the search agent together, which makes them the clearest public test of whether a co-designed stack beats an assembled one.

One engine for search, ranking and model inference over large data.

The retrieval engine the RAG boom rediscovered — sixteen years of serving infrastructure now being asked whether it is also the right substrate for agents, which is a question worth watching someone answer in public.

Agent memory that tracks what is true about a user and when it stopped being true.

The memory layer argued as a database problem rather than a prompt trick — a temporal graph with validity windows, sold into enterprise VPCs, with the open-source engine under it in plain sight.