Moss (YC F25)

I've been telling everyone Moss does sub-10ms retrieval. Time to prove it. So we did. We published full benchmark results on our GitHub repo this week. Open, reproducible, run them yourself. Here's why this mattered to me personally: 1/ I got tired of seeing infrastructure companies throw around performance numbers with no way to verify. I didn't want Moss to be one of those companies. If we're going to make the claim, the proof should be public. 2/ These aren't cherry-picked. We tested across real workloads, different dataset sizes, different query patterns. The results hold across the board. 3/ The number matters because of what it unlocks. Sub-10ms is the difference between a voice AI agent that feels instant and one that pauses. Between a search experience that's seamless and one where the user notices the wait. This was a bet we made early on. That speed would be the moat. That if we obsessed over single-digit millisecond retrieval, everything else would follow. The repo is open. The numbers speak for themselves. Link in comments.

𝗕𝘂𝗶𝗹𝗱𝗲𝗿𝘀 𝗪𝗵𝗼 𝗦𝗵𝗶𝗽 𝗩𝗼𝗶𝗰𝗲 is back. Happy hour for Voice AI builders and founders in SF. Agents, pipelines, real-time infra, if you're building on the voice stack, come through. No panels. No pitch decks. Just good conversations with people actually shipping to production. 📍 San Francisco 📅 Thursday, 16th April · 5:30 – 8:30 PM → RSVP: https://luma.com/l2fb0s4s #VoiceAI #AIAgents #VoiceAgents #SFTech #Developers

𝗧𝘆𝗽𝗲 𝗮𝘀 𝗳𝗮𝘀𝘁 𝗮𝘀 𝘆𝗼𝘂 𝗰𝗮𝗻. 𝗪𝗲'𝗹𝗹 𝗸𝗲𝗲𝗽 𝘂𝗽. We just shipped Moss semantic search plugin for VitePress docs. Runs entirely in the browser. Sub-10ms. Your user can type "how do I authenticate?" and gets the right answer, even if no page contains those exact words. Nothing sits between the keystroke and the answer. One plugin. Five minutes to set up. Try Demo: https://lnkd.in/gHuzTetJ

𝗠𝗼𝘀𝘀 𝗶𝘀 𝗼𝗳𝗳𝗶𝗰𝗶𝗮𝗹𝗹𝘆 𝗦𝗢𝗖𝟮 𝗰𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝘁 Shipping agents to enterprise usually means months of security reviews and compliance headaches. We handled the heavy lifting so you don't have to. Start building for enterprise with Moss's real-time retrieval engine now. Quick Start → https://lnkd.in/gCgC4zFY #AI #Enterprise #YC #Developers #SOC2

𝗪𝗲 𝘀𝗵𝗶𝗽𝗽𝗲𝗱 𝗮 𝘃𝗼𝗶𝗰𝗲 𝗮𝗴𝗲𝗻𝘁 𝗼𝗻 𝗼𝘂𝗿 𝗵𝗼𝗺𝗲𝗽𝗮𝗴𝗲. because the best way to explain sub-10ms retrieval is to let you talk to it. Ask it anything about Moss. It pulls from our docs. You'll hear the difference. Every other retrieval layer adds latency you can feel in a conversation. Ours doesn't. moss.dev → Start Conversation → see for yourself. #VoiceAI #SemanticSearch #YCombinator #DevTools #AIInfra

Ever tried search that works the moment you upload a doc? We built exactly that with Moss (YC F25) and LlamaIndex . Upload a PDF. Tables, columns, layout, all preserved. Search it seconds later. Locally. No database, no external services. Just sub-10ms retrieval with the exact page and source. Built on Moss. Local first. Fast by default. Try it → https://lnkd.in/g8Jy9mMW

𝗶𝗻𝗳𝗲𝗿𝗲𝗱𝗴𝗲-𝗺𝗼𝘀𝘀 𝗣𝘆𝘁𝗵𝗼𝗻 𝗦𝗗𝗞 𝘃𝟭.𝟬.𝟬𝗯𝟭𝟵 𝗶𝘀 𝗹𝗶𝘃𝗲 → comes with full python 3.14 support. → embedding generation is faster now. should make a noticeable difference in your pipelines. → You now get per-query routing metrics: local vs. cloud execution counts, latency breakdowns. Aggregated stats only, no query payloads leave your environment. Sneak Peek: Portal dashboards that bring all your analytics to the surface coming soon!