OpenAdapt

OpenAdapt automates repeated work in application interfaces that APIs can't reach. Teams and AI agent builders can turn a demonstrated task into an inspectable program, approve it for use, and check that each run produces the agreed result.

• Automation from demonstrations: A person, or an agent under human supervision, demonstrates the task; repeat runs execute the approved program without a generative-model API in the control loop.
• Result verification: Runs independently check saved results and declared unintended effects before reporting verified success, stop when verification fails, and flag uncertain outcomes for review.
• Browser, desktop and remote apps: Workflows can target browsers, native Windows, macOS and Linux applications, RDP and Citrix, with validation specific to the application and environment.
• Remote execution: A customer-controlled runner operates the visible remote client without installing software inside the remote session.

The local and self-hosted engine is open source under MIT. Raw recordings and live observations stay local by default; transferring artifacts requires explicit sanitization and approval. The local tutorial needs no account or API key, though its initial browser download requires internet access.

Commercial OpenAdapt Cloud coordinates approved metadata for customer-controlled deployments. Managed execution runs approved browser and non-sensitive workflows on OpenAdapt's servers. Paid implementation services build and maintain business workflows with agreed acceptance criteria and points where a person reviews the work.