DynamicForm Start free trial
Step 10 of 15 · Integrations · topic 8 of 8

Control · Integrations

Your integration questions, answered

These answers explain the setup process and the decisions your team needs to make. They also describe how testing, security and approval work when the connection is used for real business activity.

In short

AI proposes configuration. People review, test, approve and publish it.

Q&A

Can the platform read a provider's API document?

Yes. You can start with a written guide, an Excel field list, plain text or examples of the information sent and returned. Technical inputs are also supported: OpenAPI, Postman collections with environment settings, cURL, WSDL, and JSON or XML samples. Structured files are read directly without AI; the AI extracts relevant details from written documents. The Gap Report then shows which details are supported by evidence, which have been assumed and which are still missing, so your team knows what to check.

Which security methods are supported?

The security method is how your connection proves it is allowed to use the service. The provider specifies which method you need; a technical colleague can review that choice. DynamicForm supports API keys, Basic authentication, bearer tokens, OAuth 2.0 client credentials, password and refresh token flows, tokens obtained from another call, signed JWT, HMAC and mutual TLS. These cover keys, passwords, temporary access passes, digital signatures and certificates. Tokens are cached for reuse and refreshed once even when several requests need renewal together.

Do phones, browsers or the AI see API keys?

No. An API key or password is a credential that gives access to the provider. Credentials are entered into a protected vault as write-only values, which means the secret is not displayed back after entry. The DynamicForm server makes the provider call, so the browser and phone do not need those keys. The AI never receives keys or passwords, and personal information is masked, or hidden, before any AI call.

Can a service call back into a case?

Yes. A callback is a reply that the provider sends later, when its work is complete. The receiver checks a signature, access token or permitted network address to validate the callback, and repeated copies are processed only once. Polling is a fallback: the system checks again for the result, within a configured time limit. When the result arrives, the open web screen can update. The phone checks for the result by polling; optional reference-only push delivery requires additional setup and is switched off in the documented build.

What if documentation is incomplete?

You can still start. The Gap Report marks information as Known, Assumed or Missing and shows the evidence available for each item. The AI asks one question at a time; if you do not know an answer, it can record a safe default or prepare a question for the provider. The Assumption Ledger keeps those unconfirmed settings visible, with their risk and checking plan. High-risk assumptions must be verified before production; others must be verified or accepted by an approver.

Can the AI publish by itself, and who approves production?

No. The AI proposes settings, and people decide what to apply. Four-eyes approval means the person making the change plus one separate approver; six-eyes approval requires two separate approvers. Approvers are selected from the reporting line and receive requests in their inboxes. The maker cannot approve their own change, and the same person cannot approve twice. The final approver promotes the provider version to Production for real work. This approval process covers provider-version promotion and all reverts. Publishing a flow or saving a logical operation or provider group remains a direct action for authorised users in this release.

Can we build before the sandbox is ready?

Yes. A sandbox is the provider test system; a mock is a prepared reply that simulates the service without calling it. Mock Studio lets you prepare up to 10 request-and-response sets per API using invented data. Use them to check how screens behave when a call succeeds, fails, takes too long or is still pending. A coverage strip shows outcomes that have not been tested. You can save test cases and run them again when the connection changes.

What happens when a provider is down?

Your configured recovery rules determine what happens next. A circuit breaker stops further calls to a provider that is failing. A provider group can switch requests to an eligible backup, called failover, and return to the original provider when it recovers, called failback. Retries can wait longer between attempts instead of repeatedly calling immediately. If a call changes data, it is retried only when the provider supports idempotency, a mechanism that prevents the same request from making the change twice.

Are SOAP and GraphQL supported?

Yes, along with REST. These are different API styles: ways for systems to send requests and exchange information. REST requests can use JSON or XML data, form values or multipart content for fields and files. Your provider documentation tells your technical colleague which style and format to use. DynamicForm brings these connections into the same setup approach, so your team can map fields, test replies and choose how screens and workflows handle the results.

Can we roll back?

Yes. A published version is a saved set of integration settings. You can view earlier versions, compare them side by side and request a revert. A revert creates a new version from the earlier settings and sends it through approval. Direct provider connections can stay on a chosen version or follow compatible updates. Flows, logical operations and provider groups have one live version; their connected screens cannot remain on an older contract after a breaking change. Review the affected screens and workflows before making that update.

How long are call logs kept?

Masked call traces are kept for 90 days by default. A trace records what happened during a service request so the team can investigate a failure or slow reply, while personal information is hidden. Your team can change the retention period within the supported limits. The audit trail serves a different purpose: it records who changed settings, approved requests or performed operational actions, and can be exported to CSV for review.

Can the AI test against production?

No. An AI probe is a small test used to check an assumption, such as the format of a reply. It requires your approval and runs only against a mock or the provider sandbox, never production. The expired-token probe currently runs against mocks only. The result is saved as evidence so your team can see whether the assumption was confirmed. Production use remains subject to the review and approval process.

Build your first app free

All platform features, hosted by us for 120 days. For testing and evaluation, subject to our Fair Use Policy.

Explore DynamicForm

Compliance · Rules and calculations

Updated 16 November 2026