How a run is built
Three steps: choose nodes, fill in their settings, then execute. The screens on this page are static examples of the real editor. They are not live results.
Pick a node
The left palette lists the step types the editor actually has: HTTP requests, dataset reads and writes, AI calls, loops, branching (If/Else, Switch, Retry), and small utilities like a delay or a JWT signer. You drop them on the canvas and connect outputs to inputs. There is no separate “agent runtime” you configure off-canvas.
A workflow is a directed graph. Nodes without an incoming edge can start a run. Loops iterate over a list you already produced upstream, such as dataset rows.
- HTTP RequestCall any API
- Read DataRead rows from a Dataset
- AIOpenAI · Anthropic · Perplexity
- LoopIterate over a list
- If / ElseBranch on a condition
- Update RecordUpdate an existing row
Configure it
For an AI node you pick a provider and model from the catalog, then write the prompt. Placeholders such as Current item pull fields from the row the node is processing. The key used is the one you stored in Settings, not a ClaveFlux-hosted model.
HTTP nodes take a URL and method. Dataset nodes point at a CSV you uploaded or a Google Sheet you connected. Nothing on this screen is a live API call until you run the workflow.
Run it
A run becomes a job. You can watch progress, pause, resume, or stop. After a node finishes you can open its input and output. If you attached a dataset of examples, you can score those rows with an eval whose judge is a prompt you wrote.
The activity list below is a static mock of the jobs UI. Percentages and names are examples, not product metrics.
- Running· 17/23 rows · 73%research-and-summarize
- Complete· 5/5 nodes okresearch-and-summarize
- Eval finished· 23 rows scoredgolden@v3
Ready to try it? You will still need at least one provider key before an AI node can run.
Create an account