moving embedding into turbopuffer lets us parallelize the embedding call with other work in the query plan whole-system puffin' shaves 100+ ms off your semantic searches turbopuffer.com/blog/native-…

everyone at WarpStream sells, but @turbopuffer takes it to the next level the employees will look you in the eyes at a dinner party after just meeting you five seconds ago and ask “does your company puf? should you be puffin’?” with a completely straight face

we got something exciting for you 👀

fun things you get to see when working at @turbopuffer: today i'm fixing my code because it exceeds `i64::MAX` when summing byte size of searched namespaces in the last minute. that's 9,760,691,815,929,332,437 bytes. per minute. in a single production cluster. 9.8 exabytes :D

the first time I met CTO @nikhilbenesch was for lunch in Dec '24 in NYC, and the table at the Greek restaurant was wobbly. My grandmother Grete always carried a shim in her purse, and to pay homage to her, I always carry one too so I immediately I pulled it out at that lunch and fixed the table. that's when @nikhilbenesch knew this was it. he would join turbopuffer. I gifted him the shim. yesterday I went shimless to a restaurant in NYC. a reckless endeavour for anyone of Grete's linage, especially in NYC. fortunately, @nikhilbenesch brought his shim. he had waited 16 months for this moment. ❤️

Got a note from the @turbopuffer team that they were seeing 500 errors on a small number of our requests and they would investigate. Found the issue, patched it, and shipped it to prod. Continually impressed with their team. Can't think of another vendor has been so proactive

tpuf now supports late interaction [beta] use models like ColBERT to represent text as a set of vectors (1 per token) tpuf uses a single-vector ANN index for a fast first pass, then reranks hits using exact late interaction scoring to boost recall docs: turbopuffer.com/docs/query#l…

autoscaling is deceptively hard our indexer fleet scales nodes on job queue time. if a queue suddenly went quiet, we'd scale down too hard & new jobs could queue up waiting for nodes to claim them we tweaked the HPA signal to prevent underprovisioning → ~2x shorter queue time

Today we released Nemotron 3 Embed 8B and it reached #1 overall on RTEB 🏆 RTEB benchmarks retrieval accuracy across real-world tasks. Better retrieval gives agents more relevant context, helping improve response accuracy.

we are releasing our first open source project! puffgres is a sync service between Postgres and @turbopuffer. instead of two DB calls everywhere, you write ‘dumb’ JS transforms (i.e. tokenization, embedding) and the core handles logical replication, retries, batching, etc

turbopuffer crossed $100M run-rate in March. 19mo after $1M. Profitable & <$1M raised. Cursor・Anthropic・Notion・Cognition・Harvey・Bridgewater・Ramp・Linear・Legora・Superhuman・Atlassian・Granola We’d be nowhere without them. We work like hell to exceed their expectations.

SID-1 is an agentic search model by @SID_AI → 1.9x recall over RAG + rerank → 24x faster, 99% cheaper than GPT-5.1 trained using large-scale RL on turbopuffer at 1k+ QPS bursts over 10M+ document corpora across thousands of steps tpuf.link/sid-1

we’ve reduced query prices by up to 94%, thanks to infrastructure improvements we've made under the hood puff harder, for much less