What is an intelligent data network for AI agents?

An intelligent data network for AI agents is a single connection that lets an AI assistant or agent reach many external data sources, chooses which one to use for each request by a policy you can see, and returns each value with the record it came from. It is our name for what IKAG, short for I Know A Guy, is building; it describes a design, not an established market category.

The problem it is for

Language models are good at reasoning over what they are given. They are not, on their own, good at knowing what a company’s registered name is, what was filed this week, or which of three funds with similar names you mean. For that, an agent has to call something outside itself: a search engine, a web page, an official registry, a listings database, a social media data provider.

Each of those providers has its own API, keys, pricing and terms. Connecting an agent to one is straightforward. Connecting it to many, keeping those connections working, and knowing which one to trust for which question is a project in itself.

How it works

A request moves through the network in four steps:

1. The agent asks. It sends a request through one connection, such as an MCP server or a REST API.
2. The network works out what is needed. “Which legal entity is behind Apple?” needs company identity data from a registry.
3. It picks a source by policy. For legal identity, primary registries come first and community data only supports them. Other policies can favour price, speed or coverage, and a request can exclude sources.
4. It returns what the records establish. Each value comes back with the record behind it, conflicts stay visible, and anything the records do not settle is reported as unknown.

How it compares with what you may already use

Each approach suits different needs. These are the trade-offs as we see them; individual products vary, so check the ones you are considering.

• A single API. Simple and direct when one provider covers what you need. You build and maintain the integration, and you decide in code when to call it.
• An API marketplace. One place to find and subscribe to many APIs. You still integrate and choose each one yourself.
• A tool gateway. One key and one connection to many tools, often with per-call billing. How much each result says about where its values came from varies by gateway.
• An MCP server. A standard way to offer tools to an AI application. The model chooses between the tools it is shown, from their descriptions.
• An intelligent data network. One connection to many sources, a source chosen per request by a visible policy, and the record behind each value returned with it.

What we think matters most is the last part. When each value comes back with its record, a person can check it and an agent can weigh it. When the records do not settle something, saying “unknown” is part of the answer.

What to look for, whatever you choose

• Provenance. Does each value say which record it came from, and when that source last updated it?
• Routing you can see. Can you tell why a source was chosen, and change the policy?
• Clear costs. Can you see what a call will cost before it runs, and who bills you for it?
• Your own keys. Can you keep provider accounts you already pay for?
• Honest failure. When sources disagree or none can answer, does it say so?

What IKAG is building

IKAG is opening to early users and data partners first. Its first capability, company legal identity, answers “which legal entity is this, and where is it registered?” from the GLEIF and SEC EDGAR registries, with Wikidata as supporting evidence. You can inspect real output from it. IKAG is not affiliated with those organisations.

When IKAG opens, the AI you use will reach it through the Model Context Protocol, a REST API or TypeScript and Python SDKs. Pricing is designed around usage, and no prices have been set.