Moss is a semantic search runtime for conversational AI agents such as voice bots and copilots, where retrieval sits on the critical path of a live conversation. Instead of calling a remote vector database, it runs search and embedding inside your application's process, so a query makes no network hop and returns in single-digit milliseconds.
It has three parts. Moss Cloud handles ingestion, document embedding, storage, and distribution. An index packages your documents and their vectors as a single artifact. The runtime, embedded in your application, pulls indexes over HTTPS, holds them in memory, and serves queries locally, while document changes flow through Moss Cloud and keep the runtime in sync. The runtime embeds server-side through the Python and Node.js SDKs, or runs entirely client-side as a WebAssembly build for static sites, browser extensions, and offline-first apps.
Moss describes itself as a search runtime rather than a database: you index documents, load them, and query, without managing clusters, HNSW parameters, or sharding. End-to-end latency on 100,000 documents (embedding plus search, 750 queries, top_k of 5, on an M4 Pro MacBook Pro), with competitors using an external embedding service:
| System | P50 | P95 | P99 | Mean |
|---|---|---|---|---|
| Moss | 3.1 ms | 4.3 ms | 5.4 ms | 3.3 ms |
| Pinecone | 432.6 ms | 732.1 ms | 934.2 ms | 485.8 ms |
| Qdrant | 597.6 ms | 682.0 ms | 771.4 ms | 596.5 ms |
| ChromaDB | 351.8 ms | 423.5 ms | 538.5 ms | 358.0 ms |
Integrations cover agent frameworks (LangChain, DSPy, LlamaIndex, CrewAI, AutoGen, Haystack, Mastra, Pydantic AI, Strands Agents, Langflow), voice stacks (Pipecat, LiveKit, Vapi, ElevenLabs, Agora, TEN Framework), and web tooling (Next.js, VitePress, Vercel AI SDK, n8n). Examples include voice agents that swap indexes mid-call and a fully local voice stack with Ollama and Pipecat.
Features
- In-process search: queries run inside your application with no network hop, at single-digit-millisecond latency
- Hybrid search: semantic and keyword search in a single query
- Built-in embeddings: embedding models included, with the option to bring your own
- Metadata filtering: equality, conjunction, membership, and proximity operators
- Browser runtime: a WebAssembly SDK runs semantic search client-side with no server
- Database connectors: ingests directly from SQLite, MongoDB, MySQL, and Supabase
- SDKs and CLI: Python, TypeScript, Elixir, and C SDKs, plus a CLI for managing and querying indexes
- Framework integrations: LangChain, DSPy, LlamaIndex, CrewAI, AutoGen, Haystack, Mastra, Pydantic AI, Strands Agents, and Langflow
- Voice-agent integrations: Pipecat, LiveKit, Vapi, ElevenLabs, Agora, and TEN Framework
