Build a Workflow
A workflow is a set of steps an agent runs on its own, without needing your input at each stage. You build it once, and from then on any agent can run it with a single instruction — on demand or on a schedule.
What Is a Workflow
A workflow is a reusable sequence of steps. You define it once for something you want to happen repeatedly, and the agent executes the same sequence every time instead of you walking it through each stage.
It can be simple or multi-step:
Simple — Check a set of websites for updates each day and send you a digest in Slack.
Multi-step — Pull the day's email, extract every invoice, run OCR on each one, file them into the right Google Drive folders, and push them into your bookkeeping platform.
Either way, the workflow owns the wiring between steps. You do not repeat yourself, and the result is the same every run.
For the deeper model — nodes, sessions, publishing, and remixing — see Workflows — Core Concepts.
When to Use a Workflow
If you are going to do something more than two or three times and you want it to happen the same way each time, a workflow is the way to go. You do not have to repeat yourself, you can track every run, and the output stays consistent.
Build a workflow when the same multi-step process repeats, order matters, you want a predictable result, or you want to share the recipe.
Skip the workflow when you only need a single tool, the steps change every time, or you are still figuring out the right sequence. Let the agent pick tools freestyle until the pattern stabilizes.
How to Build a Workflow
There are three ways to build a workflow. They all produce the same thing — a named, runnable recipe — so pick the one that fits how you like to work.
If you have the AgentPMT MCP tool connected to an agent such as Claude or OpenClaw, you can build a workflow by just describing it. Tell the agent the outcome you want and the steps involved; it drafts the workflow for you. See Connect An Agent to get MCP set up first.
Open the Workflow Builder from your dashboard and use the in-builder chat. Describe the workflow in plain language and the builder assembles the nodes for you. Review and adjust before saving.
Open the Workflow Builder and assemble the workflow yourself — drop tool nodes onto the canvas, connect them in order, and configure each node's inputs. Full steps below.
Build it on the canvas
- Open the Workflow Builder
Go to the Workflow Builder. The canvas is empty by default.
- Name the workflow and write the outcome in one sentence
Examples: "Search the web for a topic and summarize the top three results into bullet points." The name and description are what your agent reads to decide whether to run this workflow, so write them for an agent reader, not a human.
- Drop tool nodes onto the canvas
Use the sidebar to search the marketplace. Click a tool to add it as a node. Each node is one tool call.
- Connect nodes in execution order
Draw a line from each node's output to the next node's input. The connection defines both order and data flow — the next node sees the previous node's output.
- Configure each node's inputs
Click a node to open its config panel. Pass literal values, reference outputs from earlier nodes, or pull values from the workflow's inputs.
- Publish or keep private
Private keeps the workflow in your account only. Public gives it a marketplace page so others can discover and run it.
Write the description for an agent reader
Agents pick workflows by reading the name and description. Describe the outcome (what the user gets at the end), the trigger (when this is the right workflow to use), and any inputs the user must supply.
Start with two or three steps
Most useful workflows are short. Long workflows are harder to debug, more expensive per run, and more brittle when one tool changes. If you reach six steps, ask whether two of them belong in their own workflow.
Make every node's input explicit
Do not rely on the agent to "figure out" what to pass. Reference previous-node outputs by field, set sensible defaults, and validate at the node where bad input enters — not three steps downstream.
Test each tool individually first
Before wiring tools together, run each one alone with the same inputs you will pass in the workflow. If a tool needs credentials, bind them on the Agent Group before testing.
How to Test a Workflow
You can test a workflow from three places:
- The builder — Run the workflow from the canvas with a sample input. Each node lights up as it executes, and any error shows on the node that failed. Fix it and re-run.
- Your dashboard — Run the workflow against a hosted agent to confirm it works end to end the way an agent will actually invoke it.
- An outside agent — Trigger it from Claude, OpenClaw, or any MCP-connected client to confirm the agent picks it up and completes it.
Run a Workflow on a Schedule
There is a scheduler in your dashboard that runs workflows on repeat automatically. You set the schedule and AgentPMT boots the workflow at the right time.
Use Control Center -> Workflows -> Schedules when a workflow should run on a calendar or interval. Schedules are attached to an Agent Group and can only target workflows that the group is allowed to access.
For the two schedule execution modes (bring-your-own agent vs. AgentPMT managed run), credit requirements, and the BYO claim API, see Using Workflows and Tools.