Skip to Content
Welcome to 10ex Docs — explore Tutorials, Guides, Reference, Concepts, Use Cases, and the Agent Store.
GuidesWorkflows
How-to

Workflows

Workflows are the orchestration layer above agents. An agent runs end-to-end on its own. A workflow chains multiple steps (agents, HTTP calls, custom code, conditionals) into a graph that runs on a trigger.

Three product surfaces:

SurfaceRoutePurpose
My Workflows/flows/listWhat’s running and recent runs
Template Gallery/flows/templatesVetted templates ready to install
Create New/flows/createAuthor a custom flow

For the runtime model and trigger types, see Concepts → Flows.

My Workflows

My Workflows page with counters Total, Active, Scheduled, Runs Today and active workflow tiles
Workflows, My Workflows. Each tile is one workflow with status, created date, and Edit or More actions.

Counters at the top (Total Workflows, Active, Scheduled, Runs Today) give a workspace-level read at a glance. The tile list below shows each workflow with name, description, status, created date, and an Edit Workflow button.

Top-of-page actions: Templates, Request Custom, Import, Create New Workflow.

Templates are pre-built workflows you can install with one click. They’re versioned, vetted internally, and parameterised. Installing copies the graph into your workspace and prompts for inputs (which segment, which connector, and so on).

The gallery is empty in fresh workspaces. As 10ex publishes templates they appear here. Search and filter by category.

How to create a new workflow

Authoring a workflow is a 2-step start, then a visual canvas.

Step 1. Name and describe

Fill Workflow Name and Description. The description shows up in the My Workflows tile and is how you’ll find it later.

Create New Workflow form with Name and Description fields and a Continue button
Step 1, name and describe. Step 2 (input fields) follows; the visual canvas opens after.

Step 2. Define input fields

Inputs are the parameters a run takes, the same way an agent’s Configure panel exposes inputs. Define typed fields here so your workflow can be triggered with structured data (segment id, target market, budget, and so on).

The canvas

Once created, you land in the visual builder.

ICP Lead Generation workflow with four nodes: Build Apollo Search Query CREW, Fetch Leads HTTP, Enrich Hunter PYTHON, Save to CRM CONNECTOR
An existing ICP Lead Generation workflow. Four mixed-type nodes wired left to right.

Top-bar actions:

  • History: past runs of this workflow with status and inputs.
  • Extend: add steps after the last node.
  • Export: download the workflow as JSON.
  • Schema: edit the inputs and outputs definition.
  • Save: persist edits.
  • Run: manual run with the current inputs.

The right pane is the Library. Drag a node onto the canvas to add it.

The 9 node categories

The Library groups available nodes:

CategoryCountWhat’s inside
Triggers2manual_trigger (UI or API run), cron_trigger (scheduled)
Internal Actions3Built-in 10ex actions: segment ops, lead ops, knowledge writes
AI Crews9Every persona that can be embedded as a step (Marcus, Nova, Sora, Maya, Juno, Atlas, Iris, Marcus/Prospector, Kai)
Flow Control7Branch, switch, loop, parallel, wait, end, error-handler
Data Transform3Map, filter, aggregate
External App Steps2 appsSlack post, Google Sheet write. More apps land alongside connectors.
CRM Connectors5HubSpot push and pull, Salesforce push and pull, 10ex CRM write
HTTP Requests1Generic outbound HTTP call
Custom Code1Inline Python (sandboxed)

A real workflow mixes types. ICP Lead Generation above is CREW → HTTP → PYTHON → CONNECTOR. Marcus/Prospector builds a search query, Apollo fetches leads, custom Python enriches with Hunter.io, and the CRM connector saves results.

Triggers

Two trigger types:

  • manual_trigger: runs on-demand (button click, API call, MCP tool).
  • cron_trigger: runs on a cron schedule.

A workflow can have multiple triggers. One cron plus one manual is a common pattern: run nightly, but also let a human kick it off.

Run history

Click the History button on the builder, or click into a tile from My Workflows. Each run stores:

  • Trigger source (manual or cron)
  • Inputs payload
  • Status per node
  • Output payload
  • Total credits spent (if any AI Crew nodes ran)
  • Total wall time

Failed runs show the failing node highlighted. Click in to see the error and re-run with adjusted inputs.

When to build a workflow vs hire a single agent

Pick the right primitive:

  • Hire an agent when one persona (with its sub-agents) covers the motion end-to-end. Most use cases.
  • Build a workflow when you need conditional branching, multiple agents in series, custom data transforms, or scheduled execution. Examples: nightly lead enrichment, multi-segment ad creative variants, lead-routing rules, post-webinar follow-up automation.

Workflows are heavier but composable. Agents are lighter but opinionated.

MCP equivalents

MCP prompt

list_workflow_templates() // available templates from the gallery

MCP prompt

create_workflow_from_template({ template_id, params })

MCP prompt

trigger_workflow({ workflow_id, inputs })

Common questions

Can I share a workflow with another workspace? Use Export to download JSON, then Import in the target workspace. Connector references rebind on import.

What happens if a node fails halfway through? The run stops at the failing node. Add a Flow Control error-handler upstream to catch and route around it.

Can I version a workflow? Edits create a new version automatically. Run history is tied to the version that ran, so you can always trace a result back to the exact graph that produced it.

Last updated on
Unified ChatKnowledge