Credit to the 'just nest more folders' school of agent memory — infinitely organized, and somehow never where the answer is.

We'll hear from @aHev an 1.5 hours on his fun work with agentic memory at Retrieval Augmented Gathering, hope you can make it! maven.com/lls/f27953

native reranking is now in private beta rerank query results using state-of-the-art reranker models, including Jev, without a third-party API call contact us for access

Agent memory retrieval is rarely an unfiltered top-K. Usually: this user, last 30 days, this topic. 1M turns with 384-dim embeddings in chDB: that WHERE prunes to 178 rows. Exact cosineDistance over them, cold Python process: 14ms. Ranking all 1M: 250ms. No ANN index needed.

Sharp find in this RAG retriever benchmark: allow_experimental_vector_similarity_index must be set at query time too, or ClickHouse silently falls back to a full scan. index_granularity 8192 -> 128 took p50 from 43ms to 11.5ms. qiita.com/kenimo49/items/c5c… @kenimo49

Quick Superlogical demo to end the week. This demo is just of the basic functionality, but I want to highlight just how FAST it is. There's a WHOLE lot more functionality already present and coming but I wanted to keep the demo short. But seriously, check out that speed. I'm showing the macOS app here and I get asked a lot so let me just say up front: its not macOS only. We're still figuring out exactly what platforms will be stable enough for the initial public release, but we're working on others. In particular, the web interface is very functional but I'll demo that another time (or someone else will). This is also all showing off the architecture I previously talked about for terminal multiplexing which is significantly different from other mainstream multiplexers: this is all a custom binary protocol where the server is maintaining N replica distributed terminal state machines. It works! Tons of hard work going into this by the team @almonk @dizzyup @pearkes. Hope we can bring you something you can use soon. ❤️ Happy to answer questions in the replies.

there’s a post I keep hesitating to write, but have mentioned offhand in pubs, podcasts, probably some tweets as well. about very senior engineers who have painted themselves into a corner by being absolutist anti-ai, and have rationalised it deeply, and made/making themselves completely unemployable. it’s super tragic to see. we need to give them some kind of offramp, but probably won’t because what exactly is the benefit to the olive branch presenting person? and yeah I’ll say it, being on bluesky is corrupting for this reason. you’ll have the constant applause from others who are in the same position, and feel completely validated in your decision, and fade into the nothingness. it’s like these people forgot what being a technologist really means, what the job is, how to be practical and adjust to a changing world. corrupted by pithy ideology that doesn’t survive contact with reality. ps - don’t use this as an excuse to be cruel to them, I see the thigh slapping self congratulatory types as well who enjoy seeing this train wreck play out. (replies limited for this one, I only discuss this with my friends in person sorry.)

When vector search returns poor results, check whether the query and documents were embedded using the same model and preprocessing steps before assuming the search algorithm itself is broken. Mismatched embedding pipelines are a far more common cause than the underlying search logic.