Moss

What it does

Moss provides a sub-10ms retrieval engine designed for real-time AI systems, particularly voice agents and copilots. Unlike traditional vector databases that introduce latency through network hops, Moss runs search directly where the AI executes—browser, edge, device, or cloud. It eliminates the need for external retrieval layers, enabling instant context retrieval for conversational experiences.

Who it is for

Moss is built for developers and teams building production AI systems where latency directly impacts user experience. This includes voice AI applications, copilots, real-time knowledge search, and on-device/edge AI. The website lists integrations with voice AI platforms (LiveKit, Pipecat, ElevenLabs), LLM frameworks (LangChain, DSPy), and frontend AI tools (Vercel AI SDK, Next.js). Trust logos include Grammarly, HubSpot, Microsoft, EPAM, and several universities (Berkeley, CMU, Stanford, UC Davis), as well as Zomato and Podium.

Why it matters

Retrieval latency is a critical bottleneck in real-time AI systems. Moss claims end-to-end retrieval latency under 10ms, up to 100x faster than vector databases like Pinecone, Qdrant, and ChromaDB (based on benchmarks on 100K documents). By running locally or at the edge, Moss reduces infrastructure overhead and enables offline-capable, privacy-preserving search. This matters for voice agents that need to respond instantly without lag, and for any AI system where milliseconds affect user experience.

Launch signal

Moss was featured on Hacker News as a Show HN post titled "We put voice agent on our website, learned retrieval isn't bottleneck." The company is backed by Y Combinator (logo displayed on the website). The product appears to be in active development with a public portal for latency testing and a founding agent offering. The website includes a live demo where users can run retrieval queries.

Brand and naming

The name "Moss" suggests something that grows quietly and pervasively, aligning with the product's promise of seamless, low-overhead integration. The brand positions itself as a fundamental infrastructure layer—"replace your vector database"—rather than a standalone application. The tagline "Built for Production AI Systems" and the emphasis on benchmarks and integrations convey a developer-first, performance-obsessed identity.