Show it once. It runs forever. On your premises.
OpenAdapt compiles a recorded demonstration into a self-healing automation that halts rather than guessing. It's open source and auditable, and it runs entirely on your own machines.
Built for the repetitive work trapped in the systems no integration ever reached — legacy EMRs, Citrix desktops, and the line-of-business apps your team still runs by hand. Where a demonstration is the only way in, that's where OpenAdapt works.
Process
How it works
A demonstration compiler: one recording in, a runnable automation out.
- 1.0
Record
Do the task once while OpenAdapt captures screens and inputs.
Record — real footage · driving a live OpenEMR instance - 2.0
Compile
The demonstration becomes an editable script: visual anchors, per-step assertions, parameters.
Compile — crafted annotation over a real recorded frame - 3.0
Run
Compiled replay in milliseconds, locally, with no per-run model calls.
Run — real footage · driving a live OpenEMR instance - 4.0
Self-heal
When the UI drifts, a fallback ladder finds the target and proposes the fix as a diff.
Self-heal — real footage · driving our MockMed demo app - 5.0
Audit
Every run produces an illustrated report: what ran, what it saw, what changed.
Audit — real footage · driving our MockMed demo app
Real EMR · measured 2026-07-08
Same 18-step task on the live OpenEMR demo, one success check for both. Both finished every run; the difference is what each run costs you.
Safety
It halts instead of guessing.
When the UI drifts and a step gets ambiguous, OpenAdapt stops and flags a person rather than writing to the wrong record. We built an adversarial test suite to measure how often it could still pick the wrong target, put our own engine through seven rounds of it, and publish every result — including what it gets wrong today.
Industries
Built for regulated back-offices
Record the workflow once and OpenAdapt compiles it into an automation your team can review, run, and audit, entirely on your own machines. No brittle selectors to hand-author. No per-run model costs.
Built-in PII/PHI scrubbing keeps your sensitive data safe.
Healthcare clinics
- Referral and fax intake, EMR data entry and extraction; web EMRs today, desktop and VDI EMRs on the roadmap. PHI handling stays fully local.
Mortgage & lending ops
- Loan-file data extraction and entry; desktop LOS like Encompass on the roadmap. Borrower data stays in your environment.
Let us build for you
If OpenAdapt doesn't fully automate your workflow out of the box, we'll work with you to fix that.
Start IntakeRunning the same workflow hundreds of times on software you can't API into?
We're taking a small number of healthcare and lending pilots. OpenAdapt v1 is open source and under active development.
Book a pilot →Pricing
Priced on the outcome, not the click
The engine is open source and free. When you need it in production, inference is bundled in — no per-step, per-seat, or per-model-call metering. On-prem keeps your data in your building.
Open Source
For builders and self-hosters who want it running on their own machines.
- Record → compile → replay, fully local
- Self-healing scripts, reviewable as diffs
- Self-host for full auditability
- The engine is MIT and always will be
pip install openadapt
Hosted
For teams without on-prem hardware who want a managed cloud runner.
- Fully managed cloud runner, nothing for you to run
- Up to 10,000 workflow runs a month
- Hosted inference included at no extra cost
- No per-step or per-seat billing, no surprise charges
Enterprise
For regulated teams — healthcare, lending, and other compliance-bound back-offices.
- On-premises deployment — PHI never leaves the building
- Inference runs on your hardware at zero metered cost
- Paid pilot → annual platform license
- BAA, SOC 2, credential vault, and audit trail
- Self-healing and fleet management included
Guided setup
Not sure where to start? We'll build your first workflow with you.
Pick one well-scoped task. We record it and compile it with you, then get it running end to end with effect verification, in days. A flat fee that's credited toward your pilot if you move forward. A low-risk way to see it work on your real workflow before you commit.
Setup pricing is a starting figure and confirmed before we begin.
We price the workflow outcome and the compliance, and bundle the inference. No per-step, per-seat, or per-model-call charges anywhere.
Open source
MIT licensed. Install it and read the code.
OpenAdapt is an open-source demonstration compiler for desktop automation. Record a workflow once and it compiles into a self-healing automation that runs on your own machines with no per-run model calls.
Install in 30 seconds
One command installs everything you need. No Python setup required.
$curl -fsSL https://openadapt.ai/install.sh | sh
The full loop, in five commands
Try record → compile → replay → heal right now against the bundled demo app. No account, no cloud, nothing leaves your machine.
pip install openadaptInstall the demonstration compileropenadapt flow demo-record --out recRecord a demonstrationopenadapt flow compile rec --out bundle --name my-taskCompile it into an editable workflowopenadapt flow replay bundleReplay: local, with zero model callsopenadapt flow replay bundle --drift themeDrift the UI and watch it heal itselfEvery replay writes an illustrated run report: what ran, what it saw, what healed. Source on GitHub. Running this on your own desktop workflows in a regulated environment? Book a demo.
PyPI Download Trends
Historical download statistics for OpenAdapt packages
FAQ
Frequently asked questions
- What is OpenAdapt?
- OpenAdapt is an open-source demonstration compiler for desktop automation. You record yourself doing a task once, and OpenAdapt compiles that recording into a self-healing script that runs on your own machines. Healthy runs make no cloud model calls, so each run costs nothing.
- How is OpenAdapt different from RPA tools like UiPath?
- Traditional RPA makes you hand-author brittle selectors and flowcharts, and the bot breaks when the UI changes. OpenAdapt compiles the automation from a demonstration instead, with no selector authoring, and when the UI drifts it heals itself and proposes the fix as a reviewable diff. It also runs entirely on your machines.
- How is OpenAdapt different from AI computer-use agents?
- Computer-use agents re-reason through your task with a large model on every run, so they're slow, non-deterministic, and bill you per run. OpenAdapt compiles the task once and replays it locally for free, with no per-run model calls. A model is only invoked to heal the script when the UI drifts.
- Does my data leave my machines?
- No. OpenAdapt is local-first by architecture: recordings, compiled scripts, and replays all stay on your own infrastructure. Nothing is uploaded to run an automation. PII/PHI scrubbing tooling is included for teams that need to sanitize captured data before anyone (human or model) sees it.
- Is OpenAdapt free?
- Yes — OpenAdapt is MIT-licensed open source, free to use and modify. For regulated deployments in healthcare and lending we run commercial pilots with hands-on support: we help compile your first workflows, set up review, and keep the automations healthy. Book a demo to talk about a pilot.
- What software does OpenAdapt work with?
- OpenAdapt works from pixels and inputs rather than APIs or browser internals. Today it runs web apps, including web EMRs. Because the runtime is vision-based, the same approach extends to desktop applications and software delivered over VDI or RDP as adapters, not rewrites — and those adapters are in progress. The aim is to reach the desktop EMRs and loan origination systems that cloud automation tools can't, without an API integration project.
Book
Book a 15-minute automation fit call
Tell us your highest-friction workflow and we'll map what can be automated first.
Register for updates (we promise not to spam)