Novalink

Documentation

Generate AI agent workflows from a description

Updated September 19, 2026

Instead of starting from an empty canvas, describe what the workflow should do. Claude drafts the whole graph against the live list of node types, Novalink validates it, and you review the result before anything is saved.

Start a draft

On Workflows, open the arrow next to New workflow and choose Generate with AI. Describe the trigger, the steps and where the result should go:

A good description
When a GitHub issue is labelled urgent, have an agent summarise it
and decide the team, then post the summary to #support in Slack.

Pick Sonnet 5 for better graphs or Haiku 4.5 for faster, cheaper drafts. Generation runs on Novalink's own key, so you do not need a credential to try it.

What makes the draft valid

The draft goes through the same validation as a workflow you build yourself: node configs match their schemas, templates only read upstream nodes, and every node is reachable from the trigger. If anything is wrong, the problems go back to Claude for one repair round, and whatever is still open is listed in the preview so nothing is hidden.

Your apps and APIs are in the draft

Claude knows the actions of every connector and the endpoints of the APIs you added, by their real ids, so a request that mentions GitHub or your billing API becomes the right node. When exactly one connection fits an app, it is filled in for you; otherwise you pick it after creating. Keys are never part of a draft.

Refine before you create

The preview shows the graph on a canvas. Ask for a change in plain language, such as "also create a Linear issue for bugs", and Claude revises the whole graph. When it looks right, choose Create and open to save it as a new workflow and continue in the editor.

Limits

Each account can generate 10 drafts in any 10 minutes. A draft is a starting point: add credentials to agent nodes, test a node or two, then run it.

Build it on the canvas

Create a free account, describe the workflow or wire it yourself, and run it in the browser.