Introducing Data Gateway: One Priced Data Catalogue for Every Agent

Most agent work starts with a lookup

A large share of the work people hand to agents begins with a fact the agent does not have. Who leads this company and how much has it raised? What did that stock report last quarter, and what do analysts say now? Is this counterparty on a sanctions list? What does a senior engineer earn in Austin? What are buyers saying about the competitor on software review sites? What did that importer ship last year?

Each of those answers sits behind a different public source with its own API, its own key and its own billing. An agent either scrapes pages through a browser, which is slow and breaks, or guesses. Today we are launching Data Gateway: one priced catalogue of read-only data tools across 18 domains, which every agent in Naive Managed Agents can search and run and which developers can call directly. Social data, the catalogue of public social media, launches with it as its social domain. The Data primitive page summarizes the surface; the API reference has every field.

Eighteen domains, one catalogue

Data Gateway holds about 1,900 ready endpoints from 51 sources today. Each tool belongs to one or more domains, and every search can be narrowed to one.

Tools are organized by the kind of data they return, such as real-estate listings, company earnings or software reviews. No connected account is needed for any of them: the agent runs them through Managed Agents and you are billed in one place. Use it for lawful purposes only, and respect privacy laws and each source's terms, especially for data about people.

The catalogue grows by itself. When a new endpoint becomes available upstream in a domain we cover, it appears in search with no change on your side, so the set an agent can reach next month is larger than the one it reaches today.

Three tools on every agent

A catalogue of 1,900 endpoints does not fit in a prompt, and it does not have to. Every agent gets three built-in tools that let it find what it needs when it needs it.

The loop is the one a person would follow. Asked about a company's last quarter, the agent searches for "earnings results" in the finance domain, reads the description of the best match to see which inputs it takes and what it costs, then runs it with the ticker. None of the three changes anything anywhere: every tool in the catalogue reads. Because search and describe are free, an agent can look around as much as it likes and only pays when it fetches data.

Calling it yourself

The same catalogue is a primitive you can call without an agent, from a script, a backend job or a notebook. The REST routes are three.

GET  /v1/data/tools?query=earnings%20results&domain=finance
GET  /v1/data/tools/:id
POST /v1/data/runs   { "tool_id": "…", "query": { … } }

The TypeScript SDK is one method per route.

const { data: tools } = await vetta.data.searchTools({
  query: "earnings results",
  domain: "finance",
});
 
const tool = await vetta.data.getTool(tools[0].id);
// tool.input is the JSON Schema for path, query and body; tool.price is what a run costs
 
const run = await vetta.data.run({
  tool_id: tool.id,
  query: { /* the inputs tool.input asks for */ },
});
// run.data is the result; run.cost_micro_usd is what was billed

The CLI covers the same three steps, which makes it easy to try an endpoint before wiring it into code.

vetta data search "earnings results" --domain finance
vetta data describe <tool_id>
vetta data run <tool_id> --query key=value

Tool ids are opaque: pass back the one search gave you, unchanged. Details: API, SDK, CLI, and the agent tools in the tools reference.

The social data routes, SDK methods and CLI commands are a shortcut to the social domain of the same catalogue: the same tools under their own names.

Try it in the dashboard

Under Settings → Data the dashboard has the whole catalogue, with a chip for each domain: search it, open a tool to see its inputs and price, fill in the inputs and run it. The data comes back on the page next to what the run cost. That is the quickest way to check that an endpoint returns what you need before an agent depends on it.

What it costs

Searching the catalogue and describing a tool are free. A run costs the price the tool lists, which search and describe both show before you run anything. That listed price is the price you pay. A typical call is about $0.002 and most are under $0.02; a few company checks cost up to about $1.24 per call, and some tools are priced per result returned, and say so in their price.

Runs are billed from the same prepaid credits as everything else, on the search spend component of an agent's breakdown, beside web search and fetch. A failed run is normally not charged.

Because some tools cost more than a cent, and a per-result tool can return many results, each run is capped at 25 results and admitted at the most it can cost before anything is bought:

• Over the API, your balance must cover the listed price of one call, or of 30 results for a per-result tool (some sources bill results in blocks of 10). If it does not, the request is refused with insufficient_credits and nothing runs.
• In an agent session, the agent takes a hold against the session's budget before it runs the tool: a per-call tool's price, or up to 30 results' worth. The hold is settled to what the run actually cost, and by default the agent asks you before each run.
• List inputs count too. A per-result tool that takes a list, such as usernames where each one is a result, accepts at most 25 items; a longer list is refused with validation_failed before anything runs.

What it is, and what it is not

Being clear about the edges:

• No logins. Data Gateway never signs in as you or anyone else. It cannot see private accounts or anything behind a login.
• Read-only. Every tool reads. Nothing is posted, sent, bought or changed on any site. Publishing to social networks is the Social primitive.
• On-demand reads, not a feed. Each run answers one question when you ask it. To watch a price, a filing or a hashtag over time, run it on a schedule with a cron deployment.

Get started

Every agent already has the three tools; ask one to research a company, its funding and its reviews, and watch it search the catalogue. To call it yourself, open Settings → Data in the dashboard, or run vetta data search. The full surface is in the Data reference.