VaraHelp Center
Contents — Tap to expand

From AI Plan to Runnable Experiment

You've discussed an experiment plan with ChatGPT, DeepSeek, or Claude. The next step isn't to copy it into a document, but to turn it into a workflow you can run, with linked parameters and failure review. Vara connects it in five steps and avoids cold-start data entry.

Five steps

  1. 1. Copy the import prompt in Vara

    When creating an experiment in Vara, open 'Copy a prompt for your AI' under the 'Paste text' card and copy the import prompt. This prompt makes your AI output the plan in a structure Vara can parse.

  2. 2. Have your AI generate in that format

    Paste the prompt into your existing ChatGPT, DeepSeek, or Claude chat along with your experiment idea. The AI produces a plan with structured steps and parameters.

  3. 3. Paste back into Vara and parse

    Copy the AI's output, paste it back into Vara, and click parse. Vara turns it into a parameterized step tree, not plain text. This step avoids cold-start data entry.

  4. 4. Check the source-confidence labels

    Vara labels each field with source confidence: explicit, partial, inferred, or missing. Focus on the 'inferred' and 'missing' fields, fill or fix them, without re-typing everything.

  5. 5. Run and record

    After confirming, run the steps with timers and log results and deviations. Change one parameter later and dependent downstream steps are flagged for recalculation; if a run fails, AI can help review which step likely went wrong.

FAQ

Why not just save the AI's answer in a document?+
A document is just text: changing one parameter won't recalculate downstream steps, failures aren't diagnosed, and iterations aren't tracked. Vara turns the AI's plan into an executable, reproducible workflow, which a document cannot do.
Which AIs are supported?+
Any chat AI that generates text works, including ChatGPT, DeepSeek, Claude, Doubao, Kimi, and ERNIE. Send Vara's import prompt to whichever AI you discussed the plan in.
Won't the AI make up parameters?+
Vara labels every AI-extracted field with source confidence so you can see at a glance what is explicit versus inferred, and focus on the inferred ones. Calculations between parameters run on a deterministic engine, not AI guessing.