Runavelo turns a natural-language goal into explicit automation commands. Review the inputs, variables, loops, conditions, interface targets, and outputs before the workflow becomes a repeatable part of your work.
Code or temporary computer-use actions can prove that a task is possible. Long-lived automation needs an artifact a user can inspect, test, revise, and run again.
A user should not need to regenerate an entire solution to change one browser, file path, timeout, selector, or output field. Runavelo exposes individual commands and their parameters. Control flow is visible as sequence, condition, loop, and subworkflow structure.
This matters after the first demo. A website may change one attribute while the rest of the flow remains correct. A workbook may move. A slow application may need a longer readiness condition. Precise editing reduces the surface area of each repair.
The assistant can also teach the editor. Ask how to build a nested loop, where to save an output variable, or how to drag a command into the correct scope. The answer can use the same command vocabulary shown in the product rather than giving generic programming advice.
After review, ordinary workflow commands execute according to the saved logic. A browser click, spreadsheet write, condition, or loop does not call a model merely because AI helped create it. Model usage remains explicit: open the assistant for design or repair, or add a model-powered command when run-time semantic judgment is genuinely required.
The result is easier to estimate and audit than a system that sends every screenshot back to a model for another open-ended decision. It also keeps model availability separate from already-defined mechanical work.
Complete Runavelo editions can combine Windows applications, browser sessions, spreadsheets, files, APIs, Python, JavaScript, and connected Android devices in the same visual model. Data does not need to be handed from one unrelated automation editor to another just because the process crosses an application boundary.
Available commands depend on the installed edition. The open-source desktop repository provides a defined subset covering CDP web interaction, spreadsheets, basic control flow, and AI building, repair, and teaching.
Runavelo can provide the assistant with the current workflow description, command parameters, runtime logs, and error stack. This makes questions concrete: Did the failing selector contain a page-specific label? Did a subworkflow receive the wrong value? Did the browser fail to reach a ready state?
The assistant's proposal is still reviewed as a workflow change. AI can accelerate diagnosis without turning repair into an invisible action.
Local browser and desktop commands execute on the Windows computer. Information is sent externally when a workflow intentionally calls an API or cloud service, uses a cloud OCR or CAPTCHA capability, publishes an application, or sends selected context to the AI provider configured by the user.
Use synthetic test data where possible, minimize assistant context, and review the selected provider's account and retention controls for sensitive work.
Strong candidates include paginated web data collection, spreadsheet preparation, repeated cross-application entry, scheduled reporting, file processing, and multi-step operational checks. The workflow should have a permitted data source, a measurable output, and a stopping condition.
Not every task should be automated. Keep a person in the loop for high-impact decisions, narrow account permissions, and test destructive or external-facing actions with reversible data before production use.
Runavelo generates an editable visual workflow made of supported commands, parameters, variables, loops, conditions, and subworkflows. Advanced workflows may also include explicit Python or JavaScript steps.
Ordinary approved workflow commands run without a model call. Tokens are used when you open the AI assistant or deliberately include a model-powered command.
Yes. You can edit an individual parameter, captured element, branch, loop, timeout, or output without regenerating the entire workflow.
Yes. The assistant can inspect workflow structure, command parameters, runtime logs, and error-stack evidence to explain the likely cause and propose a targeted repair.
See how captured elements, pagination, validation, logs, and spreadsheet output fit into one editable flow.