The team behind MoE

The founding team, Michael, Dean, and Dustin, have built and deployed agent harnesses, large scale retrieval systems (embeddings, rerankers, search algorithms, and caching), APIs, search and agent evals, and agent integrations. They are active contributors to the open-source AI community and have built integrations with many of the popular agentic frameworks. The following body of work grounds that claim:

• Tavily Search API: a web search engine for agents that has served over a billion all-time queries and tens of millions of queries per day, used by more than 2 million developers. The MoE team designed and built a multi-billion document index of the web with a hybrid search solution combining semantic and keyword search.
• Search Evaluations: designed and run large-scale search and agent evaluations on public benchmarks, including SimpleQA and DeepResearch Bench. (Tavily search evals, DeepResearch Bench, SimpleQA)
• Tavily Research API: a deep-research harness that ranked state-of-the-art on HuggingFace's DeepResearch Bench at launch, outperforming the deep-research offerings from OpenAI, Perplexity, Claude, and Gemini. (docs, benchmark writeup)
• Tavily developer tooling: the Tavily CLI, agent skills, MCP server (2.1 million downloads), and SDKs.
• Framework and cloud integrations: production integrations with Claude, OpenAI, Azure, Bedrock, Databricks, Snowflake, IBM, LangChain, OpenClaw, Weights and Biases, and others.
• Enterprise Engagements: advised and drove AI engineering decisions at customers including NVIDIA, Amazon, IBM, LangChain, Cohere, Groq and others.