Workflow Studio: Templates, a Composer and a Scheduler
A walkthrough of Workflow Studio: start from templates, build in the visual Composer with connector, model and code nodes, run versions, and schedule them.
Yantra Team3 min read
On this page
Some work is worth doing once. Some is worth doing every morning. Workflow Studio is for the second kind: a process with several steps that runs the same way each time, on a schedule or when you ask for it again.
Open it
Pick Workflow in the apps launcher, or open /c/<chat-id>/workflow. It is available from a brand-new chat, so you can browse before you have said anything. The studio owns the chat while it is open.
The home has views for your workflows, the templates and the scheduler, and a dialog for starting a new workflow. The address bar follows you through them: a workflow you open from the home has its own link, a refresh lands where you were, and Back returns you to the list.
Start from a template
The templates shown are the ones published to your workspace. Launch one, or start a new workflow of your own from the home. There is nothing to install for workflows themselves: the workflow engine is built in. The steps inside a workflow may still call connectors you have connected, such as Slack or GitHub.
Wire it in the Composer
The Composer tab opens the visual builder for one workflow, at /c/<chat-id>/workflow/<workflow>/v/<version>. A workflow is a graph: a Start node, an End node, and between them nodes for connectors, models and code, joined by edges. A code step can transform the output of the step before it.
Every runnable workflow also has a Workflow Configs node. It holds the project the workflow runs in and an Initial Context: shared values any step can read, such as a branch name or a template id. Values you pass when you start a run are merged on top, so one workflow can serve several cases.
Credentials stay out of the graph
A step that calls a connector refers to a stored credential, never to a pasted token. Yantra's own connectors need no credential at all. If a credential is added after the workflow was created, update the workflow so the step picks it up.
Run it and watch it
Versions are kept, so a published workflow is run by version. A run is queued, then running, then completed or failed. Ask for the status of a run, follow it live, or list recent runs filtered by workflow or status. A run can be given low, medium or high priority.
Put it on a schedule
The scheduler runs a workflow on a cadence. "Set up an automation that summarizes my inbox every morning" is the kind of request it is for. For a simple recurring briefing, such as a news digest or a market forecast, a scheduled task is lighter: see Scheduled tasks and price alerts.
From chat
- "Show me the workflow templates."
- "Set up an automation that summarizes my inbox every morning."
- "Run the lead-enrichment workflow again."
- "Build a workflow that narrates a script while generating the matching video."
From the command line
The same workflows are managed with cdecli workflow: list, get, create or update from JSON, read the inputs a run needs, run a version, check or monitor a run, list runs, clean up stale ones, and delete. A run can also execute locally with the embedded Node runner. See The cdecli command line.
What it is not
A one-off task the agent can do right now is just a chat message. Live, hands-on work in a browser is Computer. Build pipelines and deployments for your own code are CI, covered in CI: build and deploy your own code beside the chat. Questions about what automation is are answered in chat.