Creating & running pipelines
A pipeline is an ordered set of steps that run AI agents toward a goal. Steps can include agent calls, if/while control flow, and approval gates where a human must confirm before the run continues.
Create a pipeline with natural language
Describe the outcome you want:
Create a pipeline that reviews a pull request: summarize the diff, flag risks, then draft review comments.
DeeWaanAI designs the pipeline — steps, agents, and prompts — and saves it to your account. Review what it built before running it.
Create a pipeline with explicit steps
When you know exactly what you want, spell out the steps:
Create a pipeline with three steps: (1) an agent that extracts invoice fields, (2) an approval gate, (3) an agent that writes the fields to a summary table.
Run a pipeline
Run my “invoice extractor” pipeline.
Long runs execute in the background — you can close the chat and come back. Approval-gated steps pause the run until you approve:
Approve the pending step in my invoice run.
Check run status
What’s the status of my last run?
You get the current step, completed steps, and anything waiting on you.
Update a pipeline
In my invoice pipeline, make the first step also detect the currency.
Updates apply to future runs; completed runs keep the version they executed with.
Your data — always exportable
Everything you build is yours. You can export a single pipeline, or everything you own, at any time:
Export my pipeline and save it in my projects folder. Export all my DeeWaanAI data.
Exports are portable JSON. They work with your chat client’s own file tools — save locally, paste into another tool, or push to GitHub. No lock-in.