Show it once. It runs forever. On your premises.
OpenAdapt is an open-source demonstration compiler for desktop and web GUIs. Record a workflow once, and OpenAdapt compiles it into a deterministic, self-healing automation that replays locally at near-zero cost, verifies its own effects, and halts rather than guessing when the screen stops matching. No API required, no per-run model calls, and on the default path no data leaves your machine.
Every automation tool assumes an API. The systems that actually run regulated work often don't: legacy EMRs, Citrix desktops, and the internal apps a team still drives by hand. OpenAdapt learns them from a single demonstration and runs them where your data already lives.
- Website: openadapt.ai
- Docs: docs.openadapt.ai
- Discord: join the community
For workflows you run over and over, re-reasoning through every step with a large model is slow, expensive, and non-deterministic, and a wrong click writes to the wrong record. OpenAdapt takes the opposite path:
- Record once. Do the task while OpenAdapt watches your screen and your clicks.
- Compile. The recording becomes a script you can read, edit, and reuse. Each step carries a template crop, an OCR label, geometry landmarks, and postconditions derived from what the demo changed on screen.
- Replay, deterministic and $0. A resolution ladder (local match, global match, OCR, landmark geometry, then optionally a grounding model) runs healthy steps in milliseconds with no model calls on the hot path.
- Self-heal. When the UI drifts, a lower rung re-resolves the target and the fix lands back in the bundle as a reviewable diff.
- Effect-verify and halt on ambiguity. Every run confirms what it changed and leaves a step-by-step report; when the screen stops matching expectations, it halts instead of guessing, and identity-verified steps refuse to act on a low-confidence match.
pip install openadapt # CLI + demonstration compiler
openadapt flow record --url <app> --out rec # record a workflow once
openadapt flow compile rec --out bundle # compile it
openadapt flow replay bundle # run it, local, $0
openadapt flow lint bundle # report coverage gaps
openadapt flow certify bundle --policy clinical-write # enforce a safety policyopenadapt flow <verb> is the recommended path; the standalone openadapt-flow
package on PyPI behaves identically. Compiled workflows can also be emitted as
Agent Skills or MCP servers so other agents can invoke them.
We keep this honest.
- Web (headless browser) path: usable today. It runs entirely in CI with no OS permissions, and it is the reference backend for the compiler.
- Desktop, Citrix, and RDP backends: validating with design partners. These adapters are in progress, not yet production paths.
- ML and training: research. The demonstration-conditioned model work is a research line, not part of the shipping compiler.
The supported product is the compiler and the governed runtime around it:
| Repository | Role |
|---|---|
| openadapt ⭐ | Install + unified CLI (openadapt flow …) |
| openadapt-flow | Demonstration compiler + governed runtime (replay, self-heal, effect-verify, halt-on-ambiguity, policies) |
| openadapt-capture | Optional native recorder for desktop GUI events |
| openadapt-privacy | Optional PII/PHI scrubbing (Presidio-backed) |
These repositories explore whether human demonstrations can improve the accuracy of general computer-use models. They are research, not required to record, compile, or replay a workflow:
| Repository | Focus |
|---|---|
| openadapt-ml | Training and inference for multimodal GUI-action models |
| openadapt-evals | Benchmark evaluation for GUI agents |
| openadapt-retrieval | Multimodal demonstration retrieval |
| openadapt-grounding | UI element localization / grounding models |
openadapt-agent is deprecated and folding into
openadapt-flow; its runtime, safety gates, and sessions duplicate the governed runtime that now lives in the compiler.Internal tooling (dev automation, social posts, approval bot, multi-model consensus, telemetry, viewer, desktop/tray shells) lives in other
openadapt-*repositories. It supports development and operations and is not part of the compiler product.
We welcome contributions. Most product work happens in openadapt-flow and openadapt.
- Join Discord
- Pick an issue from the relevant repository
- Submit a PR (see each repository's CONTRIBUTING.md)
OpenAdapt is in active development. The web compiler path is usable today; desktop/Citrix/RDP backends are validating with design partners; the ML line is research. See the website for current benchmarks and limits.
Professional implementation and design-partner engagements are available. Contact info@openadapt.ai, or support development via GitHub Sponsors.
All OpenAdapt.AI repositories are licensed under the MIT License unless otherwise specified. See individual repository LICENSE files for details.
