Introducing Yaha Workflows

A single agent can answer a question. A process that fetches data, calls an assistant, waits for a reviewer, and posts a result needs a graph.

Yaha Workflows are visual DAG automations in the console. You draft on a canvas, publish a version, and start runs by hand, from an API, on a cron schedule, or from a webhook.

Why a canvas on the same platform

Glue scripts drift. They hold provider keys, skip audit logs, and break when someone changes a prompt. A Yaha workflow keeps LLM steps on the Gateway and can invoke a published Yaha Agent as a node.

You still see tokens, cost, and per-node status on every run.

What you can put on the graph

Public node types include:

• Triggers — manual, API, schedule, webhook
• Data and logic — set variable, transform, condition
• Integrations — HTTP request and chat-platform media
• AI — chat completion, agent invocation, WhatsApp agent
• Flow — approval, delay, output

Wire fields with expressions such as {{ workflow.input.text }} and {{ nodes.summarize.output }}. Starter templates such as Agent Chat give you a working path from trigger to agent to output.

Approvals are first-class

Drop an Approval node when a person must sign off. The run pauses, the request lands in Products → Workflows → Approvals, and approve or reject resumes the matching branch. You can loop a rejection back to an earlier agent or LLM step when the work needs another pass.

Getting started

1. Open Products → Workflows in the console.
2. Create a workflow or start from Agent Chat.
3. Bind a published agent on the agent node.
4. Save, publish, and click Run.

Guides: Workflows overview, build and run, and triggers and approvals.

Ready to try it? Sign in and publish your first workflow.