Tako 2.0 is free in Vercel AI Gateway: Feeding agents all the signal, now including the web
Key Takeaways
- Tako now searches the web alongside trillions of licensed, structured data points, returning both in a single call.
- Tako Search is free in Vercel AI Gateway for the rest of September.
- We’re relaunching our API portfolio across four primary endpoints: Search, Answer, Contents, and Agent. They’re SOTA across a variety of benchmarks.
Agents are only as good as the signal they can process. The two largest sources of external signal are: (1) the open, accessible, unstructured web, and (2) millions of (often) closed, inaccessible, structured databases. Many web search API providers have launched to deliver an agent-optimized web. We built Tako to democratize access to the latter: curating an index of authoritative, real-time, structured data sets spanning finance, economics, politics, sports, and weather. The most premium sources, delivered to your agent with per query pricing, no subscriptions or MCP required.
“Web search tool” has become a default building block in agent harnesses, and most of our customers implement Tako data search alongside web search. We realized there’s an easier solution: Tako now searches the web alongside trillions of licensed, structured data points, returning both in a single call.
To celebrate, Vercel is launching an exclusive promotion: Tako Search is free in Vercel AI Gateway for the rest of September.
Inference made intelligence a metered utility, which has unleashed explosive growth in agents. Now agents want everything to flow with the flexibility and configurability of inference — infrastructure, payments, and knowledge. Adding web search made us rethink how best to make knowledge “ergonomic for agents.” The result: we’re relaunching our API portfolio across four primary endpoints: Search, Answer, Contents, and Agent.
The New Product Family
Search retrieves a rank ordered list of web and data results for your agent to reason over.
When your agent asks, “how is Nvidia’s revenue doing?” Tako responds with (a) data results spanning estimated quarterly revenue to actual annual revenue, sourced directly from S&P Global and Fiscal.ai, and (b) web results spanning the latest earnings announcement to analyst coverage. We offer Search in Instant, Fast, and Deep modes depending on your latency, cost, freshness, and accuracy goals.
Answer adds reasoning and synthesis, constructing a definitive answer to your question grounded in the results retrieved from Search. Answer comes in the same Instant, Fast, and Deep efforts, and supports structured outputs.
Contents takes web pages and Tako cards alike and fetches their full contents for your agent. For web pages this is the full article text and for Tako data index results this is a structured export of a Tako card’s backing data, available in both CSV or JSON formats.
Agent is the culmination of our other products, built to handle complex questions and deliverables. It traverses both Tako’s knowledge graph and the web together, stitching together facts, writing code, inspecting its own work, and refining its answers along the way. Agent is a great fit for asynchronous research where complexity is high and accuracy + recall are paramount.
Search Quality & Latency
Tako is in a class of its own when benchmarked against other web search APIs. For traditional, “trivia style” web search retrieval benchmarks, Tako Search tops the field at the same latency.
View data
SimpleQA accuracy by search API
Combining web and data incurs no speed penalties — a feat made possible by Tako’s unique knowledge infrastructure we purpose built to quickly and accurately operate on structured data.
View data
Advertised search latency by API
Your agent isn’t prepping for a trivia contest. The static Wikipedia-style questions found in traditional benchmarks miss the real reasons agents need more than what was incorporated in the last model training run: access to quality, fresh data. We built VerticalRTK to fill the gap for finance, macroeconomics, sports, weather, and other realtime domains. Importantly, VerticalRTK’s ground truth changes every time you run it (e.g. “what’s Airbus’ stock price” changes constantly throughout the day), making it harder to game than static benchmarks.
View data
VerticalRTK - Research accuracy by search API
Tako’s agent pushes the state of the art forward on rigorous, domain specific research. On FinSearchComp, a public financial research benchmark, our agent produced the highest accuracy at the lowest latency across both difficulty tiers.
View data
FinSearchComp T2 accuracy and time to answer by agent
Enabling New Agentic Capabilities
Tako’s refreshed API suite gives agents a completely new set of capabilities. Now your agent can:
Conduct deep due diligence on European equities like Airbus.
Keep up to date on live sports scores.
See which RFPs the Department of Defense posted today, and get a bid in before you wake up tomorrow.
See what your competitors pay engineers on H-1Bs.
And it’s not just about access to data and the web. Tako’s architecture was purpose built to quickly and accurately operate on structured data — computing averages, calculating changes, projecting trends. With data exports, that means 85% fewer tool calls and 90% fewer tokens consumed for research questions.
Compare Spain's and Vietnam's power-sector CO2 emissions since 2000. When did each country's emissions peak, and how far below its peak is each today? How many times larger are Vietnam's emissions now than in 2000? Has Vietnam overtaken Spain — and if so, in which year?
Run timeline
Looking Ahead
Every web page and every data set is a source of potential signal for your agent. We’ll keep growing our index, and finding more efficient and powerful ways to surface that signal to your agents. We can’t wait to see what you build.
Also — we’re hiring.
About the author
Alex Rosenberg
Alex is the CEO and founder of Tako, an AI data analyst that converts natural-language questions into visual analytics and insights. His previous company, Graphiq, was acquired by Amazon to power Alexa's open-domain Q&A, and he later led the Alexa Multi-Agent team at Amazon. He has over a decade of experience building and scaling knowledge graphs and data systems for AI applications.
Ground your AI agents in real-time, structured data — web and premium sources unified in one API. Try Tako in the Playground or .