Partner program

Your users still re-key into someone else's software. Ship that last mile with a receipt.

OpenAdapt turns a recorded GUI workflow into a program you can put inside your product or run for your clients. After each run, an independent read of the system of record says whether the write landed. Pass is a signed Seal. Fail stops the run. The runtime is MIT. A partnership adds the commercial terms, the qualification method, and who supports what. Execute returns the Seal to your product.

Four partner tracks

Vertical software / OEM

Vertical SaaS vendors whose users still re-key into an EMR, a practice management system, a loan origination system, or a payer portal.

Model: Embed or white-label. The program runs as a feature of your product, under your brand if you want, on OEM terms.

Responsibility boundaries

  • Qualification: we qualify each workflow on the exact target application together, before it touches a real record. We provide the tooling and the evidence format.
  • Support: you take first-line support inside your product. We take engine-level escalation.
  • Packs: the workflows for your market are yours. The runtime and the verification contracts under them are ours, and MIT.

RCM and BPO operators

Revenue-cycle, claims, and back-office teams that run the same work every day for their own clients.

Model: Managed operations. You run the programs for your clients, on your runtime or theirs, and you keep the client.

Responsibility boundaries

  • Qualification: each client workflow is qualified on that client’s exact application and environment before production. You provide the access. We provide the method and the evidence.
  • Support: you own the client relationship and first-line support. We support the runtime and the verification.
  • Packs: the workflows you build for your clients stay yours. The runtime under them stays MIT and open to inspection.

Integration and services partners

Automation consultancies and system integrators that scope, build, and hand over automations.

Model: Services delivery. You run discovery, build, and rollout on the open runtime and the published qualification method.

Responsibility boundaries

  • Qualification: you run the qualification with your clients. We review the evidence before a go-live that writes to a real record.
  • Support: you own the engagement and first-line support. We provide second-line engine support.
  • Packs: build and reuse your own accelerators. A change to the runtime or the verification contracts goes upstream, never into a fork.

MSP and deployment partners

Managed service providers that own the customer’s infrastructure, including locked-down networks and virtual desktop estates.

Model: Managed deployment. You install and run the runtime inside the customer’s boundary, on their machines or on yours.

Responsibility boundaries

  • Qualification: environment readiness is yours, meaning identity, network, the remote client, and the update policy. Workflow qualification follows the shared method.
  • Support: you monitor the environment and roll out updates. We ship runtime releases and security advisories.
  • Packs: deployment runbooks and hardening baselines are shared working material inside the program.

Verified writes

Before your agent says done, read the record.

Your agent posts a claim into a portal where the API stops. The record is supposed to take it. When it doesn't, you find out at the denial, and someone on your floor re-checks work the agent already called done.

Sometimes the screen says saved and the record doesn't. In our test, a screen-only check accepted 54 of 90 wrong results. One out-of-band read of the record cut that to 9 of 90. If your agent checks the confirmation screen, it runs the check that accepted 54. Our technical report, August 2026, synthetic data

After the run, an independent read of the record says pass or fail, without asking the agent. Pass is a Seal, the signed receipt that the write landed. Fail stops the run, and a person decides on the record. The check runs on your machine, next to your records, and only reads. Credentials never leave it.

Watch it catch one. openadapt quickstart --break-it paints the success banner on synthetic data, then refutes the write.

Who does the work

You do. $0.
The MIT worker, openadapt-flow serve-reward, on your machine. Setup is in the docs.
We run it. $29 / month.
A runner we keep on, an API key your agent calls, and a dashboard with a verify URL for each run, at app.openadapt.ai.
We do it with you.
For the first design partners, we wire the record read into one of your workflows, paid and scoped. This path is new. It starts with a conversation.

We test the check before we trust it, against a corpus of runs that look finished and aren't. Today that corpus is synthetic, so each receipt is certified against our synthetic MockMed and ExtraDup corpus and nothing else. A production-scope certificate is not published yet.

If you train

That same receipt is the reward. The model tries the job, the record read says pass or fail, and the model learns from that read. 1 when the record is right. A screenshot is not a 1. An episode the read can't score is dropped, never paid 0.

For your engineers

TRL or verl
Import the reward function from openadapt-evals. 1 on a VERIFIED receipt. An unscored episode drops.
Environment
MockMed with the ExtraDup mutations, for Prime Intellect / verifiers. Synthetic, MIT. prime env install openadapt/openadapt-mockmed-extradup
Reward worker
openadapt-flow serve-reward. Its /v1/graders/openai answers grade(sample, item) for a loop you host. A hosted OpenAI grader can't call this worker: OpenAI documents no grader that takes a URL.

Public price list

Local $0. Hosted $29 / month. Enterprise talk to us.

Local
$0
Hosted
$29 / month
Enterprise
Talk to us

Local runs on your machine with no account. Hosted means we run the machine. Enterprise starts with a review of one workflow, one app, and one environment. Full list

Program status

You apply. We read every application ourselves.

Each partnership starts from one customer workflow that you name. We agree commercial terms per partner. A workflow that writes to a real record goes live only after qualification on the exact application and environment, with the evidence the method publishes. There is no partner portal, directory listing, or certification today. See the published qualification evidence.

Apply to the partner program

The form is the same one a customer uses. The link below presets the reason to Embed in a product. If you already have a customer workflow in view, describe it there, and we start from that.