# What to ask before you connect an agent to a data gateway > Eight questions for any agent data gateway, applied to IKAG in early access: provenance, failures, routing, cost, own keys, retention, terms. - URL: https://ikag.dev/blog/questions-before-you-connect-an-agent - Author: Ronit Ahuja, Founder, IKAG (https://ikag.dev/authors/ronit-ahuja) - Published: 2026-10-10 - Updated: 2026-10-10 - Reading time: 4 min - Publisher: IKAG ## Overview Eight questions for any agent data gateway, applied to IKAG in early access: provenance, failures, routing, cost, own keys, retention, terms. ## Key points - A data gateway gives an AI agent one connection to many outside data and tool providers. Choosing one is a trust decision as much as a technical one. - Ask where each value came from, what happens when sources fail or disagree, how providers are chosen, and what a call costs before it runs. - Ask whether you can use your own keys, what is stored and for how long, whether its numbers are dated, and on what terms the data is used. - We apply the same eight questions to IKAG below. Some are answered by its first capability; several are not answered yet, and we say which. A data gateway gives an AI agent one connection to many outside providers: search, web pages, listings, company records and more. It can save you a stack of integrations. It also sits between your agent and everything it learns, so choosing one is a trust decision. For a single provider you rely on, a direct integration may still be simpler. These are the eight questions we would ask, and at the end we answer them for IKAG as it stands today. ## 1. Where did each value come from? Not "which sources did you use", but which source supplied this specific value. Ask to see a result. Good provenance names the record behind each value, when that source last updated it, and whether other sources agreed. A receipt that shows only what a call cost tells you about your bill, not about the answer. ## 2. What happens when a source fails or sources disagree? Every provider fails sometimes. Ask whether a failed call is marked as failed, whether it is retried with another provider, and whether you are charged for it. When two sources disagree, the result should say so and show which one was preferred and why, not quietly pick one. And when the sources do not settle a question, the result should say "unknown", not guess. ## 3. How is a provider chosen, and can I change it? Behind one connection, something decides which provider answers. Ask what that decision is based on: authority, price, speed, coverage, or several providers at once for comparison. You should be able to see the choice and set the policy, including excluding a provider or keeping to one you trust. ## 4. What will a call cost before it runs? Ask whether you can see the cost of a call before it runs, and set a ceiling so an agent stuck in a loop cannot run up a bill. Usage pricing only helps if the price is visible before you commit. ## 5. Can I use my own keys? If you already pay a provider, you may want to keep that plan and route calls through your own account. Ask how calls made with your own key are billed, by whom, and whether the gateway charges anything for its part. ## 6. What do you store, for how long, and who processes it? Requests can contain personal or commercial information. Ask what is logged, how long it is kept, where it is processed, and which subprocessors see it. If requests contain personal data, data protection law such as the UK GDPR or the EU GDPR may apply, depending on where you, your users and the gateway are; that is a question for the gateway and for your own legal advice. A security page with no answer to these questions is an answer. ## 7. Are your numbers consistent and dated? Catalogue sizes, provider counts and benchmark results are easy to inflate and quick to go stale. Ask for a dated figure and how it was counted, and check that the same number appears everywhere the gateway publishes it. If benchmarks are quoted, ask for the method and the raw data. ## 8. On what terms is the data used? Some data comes from official APIs, some from licensed feeds, and some from reading public pages. Ask which is which for the sources you rely on, whether the gateway respects robots.txt (the Robots Exclusion Protocol, RFC 9309) and site terms, and how it handles personal data found on public pages. ## The same questions, asked of IKAG IKAG is opening to early users and data partners first. Here is how it approaches each question; where something isn’t decided, we say so. | Question | How IKAG approaches it | | ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | 1. Each value's source | Every value names the record it came from, and when that source attested it, if it says. [See real output](https://ikag.dev/agents#real-result). | | 2. Failures, conflicts | Conflicts are kept as conflicts, gaps are reported as unknown with a reason, and a rate-limited source is refused, not retried silently. Whether failed calls cost anything will be published with prices. | | 3. Choosing a source | Primary registries first, community data only as support, and a request can exclude sources. | | 4. Cost before a call | Runs are quoted before they execute and carry a cost ceiling. Prices will be published before IKAG charges anyone. | | 5. Your own keys | The provider bills you under your plan, and IKAG may charge separately for its part. | | 6. Storage and retention | We will publish what the service stores, for how long and who processes it before it opens. | | 7. Numbers | We publish none yet. We will only publish figures that are real, dated and counted the same way everywhere. | | 8. Terms of data use | The first capability uses public records: GLEIF and Wikidata publish under CC0, and SEC EDGAR is public US government data. Commercial sources depend on an agreement with each provider; none is announced. | If that is the kind of gateway you want, see [real output from IKAG’s first capability](https://ikag.dev/agents). ## Questions ### What is an agent data gateway? A service between an AI agent and many outside data and tool providers, so the agent connects once instead of integrating each provider. Accounts and billing vary: some providers still need your own account or key. ### Should failed calls cost anything? That depends on the gateway’s terms, so ask. At minimum, a failed call should be clearly marked as failed on your receipt, never presented as an answer. ### Does data protection law apply to what my agent sends? It can. If requests contain personal data, laws such as the UK GDPR or the EU GDPR may apply, depending on where you, your users and the gateway are. Ask the gateway, and take legal advice for your situation. This is not legal advice. ## Sources - [ICO: UK GDPR guidance and resources](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/) - [Regulation (EU) 2016/679 (General Data Protection Regulation)](https://eur-lex.europa.eu/eli/reg/2016/679/oj) - [RFC 9309: Robots Exclusion Protocol](https://www.rfc-editor.org/rfc/rfc9309.html)