Frequently asked questions

Say a new compliance case opens in Jira. Rapidfolio fetches the customer record from Stripe, runs a KYC check, and posts the result to your compliance Slack channel. If the check comes back as "consider," the run pauses and routes to a reviewer. They approve or reject; the decision is logged with their name, the timestamp, and the data they saw at that moment. You wrote that workflow once as a procedure. Rapidfolio has run it the same way ever since. Teams use it for KYC/AML onboarding, credit underwriting, payment reconciliation, and dispute resolution — any operation where the steps are known and the trail matters.

Most AI agents interpret a goal and figure out the steps themselves at runtime. That works for tasks where improvisation is fine. It doesn't work when an agent is deciding whether to charge a card or approve a credit application. Rapidfolio runs your procedure exactly as written. The agent reads each step in order, calls the tool calls you configured with the parameters you specified, and follows the navigation you defined. It cannot invent steps, infer different API calls, or change what gets sent. The amount charged is what you wrote in the charge step — resolved from the variables in scope, not decided by the model. What you test in sandbox is what runs in production.

A procedure is a workflow document you write in the Rapidfolio editor. It contains steps organized into sections. A step can call an external service — fetch a Stripe record, run a Plaid balance check, send a Slack message — evaluate the response, branch based on the result, or pause for a human to review before continuing. Procedures are versioned. You test a version in sandbox, run named scenarios against it, and then set it as the live default when you're ready. At that point, the procedure is the complete specification of what will happen. The agent cannot deviate from it. Rolling back is one click.

Given the same inputs and the same integration responses, Rapidfolio always produces the same outputs. This is a hard architectural constraint, not a best-effort guideline. Every tool call has explicitly configured parameters — either fixed values or variable references that resolve at runtime. The agent does not decide what to pass or which API to call. A practical consequence: if a procedure charges a customer, you can open the run record six months later and see the exact amount, the exact customer ID, and the exact Stripe response. No inference happened. The record reflects what was sent.

You add a human review block anywhere in your procedure. When a run reaches it, execution pauses — indefinitely, no timeout — and your team gets notified via webhook. A reviewer opens the run in the dashboard, reads the context and instructions you wrote, and clicks Approve or Reject. Their decision goes onto the run record permanently: who made it, the exact timestamp, the data they saw, and any values they changed before approving. If they reject, the run fails, the reason is logged, and nothing else executes. For the highest-risk calls — a wire transfer, a bulk charge — you can also require approval of the resolved API parameters just before the call fires, so the reviewer sees exactly what will be sent and can edit it.

Native integrations include Stripe, Plaid, GoCardless, Slack, Jira, Linear, Xero, Alloy, Flagright, and more. Set them up once in Settings; every procedure in your workspace can use them. For everything else: Custom API connects to any REST service with a base URL and credentials. OpenAPI imports any OpenAPI spec and generates a callable action for every operation automatically. Private Connection runs your own code inside your network with no inbound ports — a lightweight SDK polls Rapidfolio for invocations and executes them locally. Rapidfolio runs alongside your existing tools. Nothing gets ripped out or replaced.

Direct API call from your backend, a built-in trigger (Stripe event, Jira issue created, Slack message, cron schedule), or manually from the dashboard. Multiple triggers can point to the same procedure. Each trigger is pinned to a specific version — releasing a new version of a procedure does not change what any trigger is running. You update triggers when you decide to, not automatically.

Three ways. Simulation mode in the editor does a dry run — data flows through steps and navigation is evaluated, but no real API calls are made. You configure mock responses per tool call to control what the simulation sees. The sandbox environment runs the procedure against real integrations in test mode: Stripe test mode, Plaid sandbox, your own staging systems. Real API behavior, real error codes, no live data touched. Scenarios are named test cases with fixed inputs and the outputs you expect. Run them before promoting a version to the API default. If a scenario fails, you catch the regression before it reaches customers.

Every execution creates a permanent run record. For each tool call, the record shows the exact parameters sent, the exact response received, how long it took, and which connection was used. Human review decisions are in the same record: who approved or rejected, the timestamp, what data they saw, and any values they changed. Run records are append-only. Rapidfolio does not delete or modify them. For a Stripe charge from six months ago, you can open the run and read the exact payload that went to Stripe and the response that came back — not a summary, the actual JSON.

No. Data processed through Rapidfolio procedures is not used to train any models. Data is encrypted at rest with AES-256 and in transit with TLS 1.3. Role-based access controls limit what each team member can see.

Primarily financial services: fintechs, lenders, payment processors, and operations teams at companies where a compliance failure has a direct regulatory cost. The common thread is that these teams need automation they can show to an auditor — not just faster, but traceable. The platform is also used in insurance, healthcare, and legal workflows where human-in-the-loop steps and documented decision trails are required by regulation or internal policy.

Yes. Rapidfolio is backed by Y Combinator.