What's New: April
This month: /research is becoming a building block for modern AI systems, alongside new agent skills, real-world benchmarks, and deeper integrations across the ecosystem.
Come hang out with us in person
Join us in San Francisco for LangChain Interrupt as we dive into the next chapter of autonomous intelligence. Look for our branding on city buses, and visit Booth 4 for exclusive swag.
We’re co-hosting several small events and sessions throughout the day, including LangSmith Fleet and Deep Research UI demos, plus a special AMA with our teammates at Nebius, who joined forces with us late last year to accelerate the transition from human-centric search to machine-centric retrieval.
Can't make it to San Francisco? You can also catch the Tavily team on the ground at IBM THINK in Boston, May 4-7.
/research as a building block? It’s the new thing.
/research started as a faster way to run deep research through one call that plans, searches, analyzes, and writes. What’s changed is how people are using it.
Instead of stitching together pipelines, developers are now treating /research as a building block. It’s being used inside agent workflows when a task needs more than a single search, and outside of them to power products and analyst-style work like due diligence, regulatory review, and market monitoring.
It’s a shift that’s equating to less orchestration and more building.
[Learn to use /research in your own system →]
Turning dynamic filtering into a reusable agent skill
In Anthropic’s Programmatic Tool Calling (PTC), Claude writes Python code that filters web results inside a cloud sandbox before they enter the model’s context window, which improves search accuracy. It works, but it’s landlocked to Anthropic’s API and a closed off environment.
By using the Tavily CLI and a lightweight skill, you can strip away that lock-in. Now, your coding agents can run dynamic filtering through Bash and Python directly. This keeps raw search data out of sight while bringing back cleaned, useful information.
On DeepSearchQA with Sonnet 4.6, this reaches 72.1% F1 vs 59.4% for PTC, while using about 3.5x fewer tokens at 12x less cost.
[Read the detailed breakdown →]
Snowflake x Tavily bring real-time web search into Snowflake Intelligence:
Tavily is now available as a web search tool inside Snowflake, letting agents pull in real-time external context, including news, regulatory updates, litigation, and leadership changes, while working with structured financial data.
That means a Financial Agent can analyze fundamentals in your workflows and immediately enrich them with live signals from the web.
[Check out Snowflake’s set-up guide →]
MongoDB + Tavily reduces hallucinations through hybrid search
MongoDB and Tavily now support hybrid search that combines MongoDB Atlas data with real-time web results in a single call.
Instead of relying on just internal data or external APIs, developers can retrieve both by mixing vector search with web search, resulting in fresh, trusted information. You can also control how many results come from each source.
This helps keep AI responses grounded, reducing hallucinations and improving the accuracy in retrieval‑augmented generation (RAG) systems.
[See why hybrid search matters to you →]
Where we’ve been
In April, the Tavily team traveled across coasts, from San Francisco for HumanX and AI DevDay down to Rio de Janeiro for ICLR.
We enjoyed chatting with our customers, hosting our booth at HumanX, and spending time in hands-on sessions at AI DevDay. At ICLR, we were happy to join our Nebius teammates, and talk about /research. Thank you to all who came out to see us.
Case Study: JetBrains x Tavily
While JetBrains was building their AI coding agent, Junie, they found it was confidently hallucinating on recent APIs.
By grounding responses with real-time web search using Tavily, Junie immediately switched to accurate, real-time web retrieval with citation, allowing developers to trust the code they’re writing.
YOUR INPUT RUNS THE ROADMAP
Having you as part of the Tavily community truly means something to us. We'd love to hear how things are going for you so we can keep showing up at our best.
WE'D LOVE TO HEAR FROM YOU
Loving Tavily? Tell the world.
Real reviews from real users carry more weight than anything we could say ourselves.
If Tavily has made your work easier, two minutes of your time could be exactly what someone else needs to find the right tool.