Workflows
Automate an entire process on one canvas, from the event that starts the work to all the decisions and actions that complete it. Every decision point is explicit, not inferred. A carrier check call, from the scheduled trigger to follow-up. Agents and Python run side by side, with tools, branches and reusable workflows connected on the same canvas.
Book a demoTriggers
Every workflow starts with a trigger, like a call, message, API request, webhook, uploaded file or a schedule you set, so work begins where it already arrives instead of waiting for someone to kick it off.
- Thu 1 Oct, 07:00Next
- Fri 2 Oct, 07:00
- Mon 5 Oct, 07:00
Core nodes
Each node does a specific job: extract information, classify it, run a calculation, call a system, or decide what happens next, automating the predictable steps.
- AI ExtractTurn unstructured inputs into fields or a defined JSON Schema.
- AI ClassifyClassify inputs into the categories your process uses.
- AI GenerateGenerate content from instructions and workflow context.
- Custom CodeRun sandboxed Python for calculations and transformations.
- File OperationsParse, convert and extract information from files.
- ConditionalsRoute work according to explicit conditions.
- LoopsRepeat steps across a count or collection.
- WebhookSend requests to APIs and receive responses.
- SchedulePause until a set time or wait before the next step.
- Module ChangeMove to another module and pass its context along.
- Call WorkflowDelegate work to a reusable workflow and use its response.
Branching and iteration
Route work down the first matching path, the last, or every one, with branches running in parallel and merging back together. Loops repeat over any list or a fixed count, tracking each item as its own run, so a batch never loses the detail.
- Running lateeta_delay > 30
- Missing PODpod_received = false
- Rate disputerate_change != 0Skipped
Leverage Enterprise Data
Query data produced by agents, search knowledge bases and pull records from connected systems, then carry those results into the next step as variables. Agents have access to the latest data before the conversation even starts.
Confirm load 48213 still arrives by shipments.eta
- load_id
- lane
- eta
- carrier
Versions, environments and publishing
Edit a draft, test it in development or staging, then publish to production, with each environment keeping its own version, credentials and endpoints. Require approval, restore any earlier version, and improve a live process without interrupting runs already in progress.
- Version 14Draft, edited 12 min agoDevelopment
- Version 13Tests passedStaging
- Version 12Live, 2,140 runs todayProduction
- Version 11Restore availableArchived
Requested by the build team. Staging tests passed.
Awaiting owner approvalRequest changesApproveApproved by the ownerPublishingFAQs
Beyond AI agent nodes, workflows use webhook nodes for outbound API calls, code blocks for deterministic transformations, Extract/Classify/Generate AI nodes for structured output outside live conversation, and third-party integration nodes connecting to 200+ external systems.
Workflows support every major trigger pattern including an inbound call, a webhook event, a scheduled run, a file upload, or an inbound message, so the same underlying workflow logic can be kicked off by a phone call, an API call, a cron schedule, or a document landing in a system, without rebuilding the flow for each entry point. This decouples what triggers a workflow from what the workflow actually does, so the same operational logic can serve multiple channels or systems.
Production is always isolated from the testing cycle: a prompt change being tested in staging cannot break a call in production, and an integration update in development cannot affect a live workflow. New versions go through an approval process modeled on a Pull Request workflow, so no change reaches production without explicit sign-off from the right stakeholders.
Yes. Nodes can execute sequentially, in parallel, or in loops, for example calling three APIs simultaneously and merging results, or iterating over a list of records, with no platform-imposed constraint on how nodes are connected or sequenced.
Yes, through reusable workflow components, a shared library maintained at the workspace level. Prompt components let agents import shared prompt logic like guardrails or identity verification instead of duplicating it. Node components package a reusable integration configuration, like an API endpoint and its authentication, that any workflow can import. Workflow components go further, packaging a complete sequence of nodes, agents, and integrations into a callable function: another workflow can invoke it, pass in data, and use the result without knowing what happens inside.
