Loading…
Workflows
Combine agents, data, documents, browser robots and human approvals in durable workflows. Draw them on a canvas or describe them in a sentence.
Capabilities
Describe the process and a runnable graph streams onto the canvas, ready to edit.
Static or AI-written reviewer messages, required roles, policy tags and timeouts from 4 to 72 hours.
Paused runs are checkpointed to Postgres and resume from the same node, on the same version, even after a restart.
Per-node retry policies with exponential backoff, plus a DLQ to inspect, resolve and replay failed runs.
Manual, schedule, webhook, SharePoint folder watch, Slack or Salesforce events, forms, chat and other workflows.
Watch every node stream status in real time: running, done, failed, waiting, retrying or skipped.
How it works
Start on a schedule, webhook, new SharePoint file, app event, form, chat message or another workflow.
Query data, extract documents, call agents and teams, drive browsers, branch and loop.
Route exceptions to the right reviewer with context, confidence and a deadline.
Notify, upload to SharePoint, append to Excel, or hand results to the next system.
Human in the loop, done properly
After a Doc Extract step, the Approval Gate switches to document-review mode: only flagged or high-risk fields, inline edits, a bounding-box overlay on the source, and auto-approval above a confidence threshold.
Built for production
Workflows are versioned and published like agents. Executions pin their version, so a run that resumes days later still follows the definition it started with.
SharePoint folder
New file trigger
01Doc Classify
Cheque · transfer · mail
02Doc Extract
Fields + confidence
03Approval Gate
Low-confidence only
04Agent
Posts and notifies
05Under the hood
The specifics your architects and security reviewers will ask for.
Works with
Every capability shares the same runtime, identity and governance.
Yes. The AI Workflow Builder drafts a runnable graph from a plain-language description, and the visual canvas makes every step editable without code.
Each node can retry with exponential backoff. Runs that still fail go to a dead-letter queue where you can inspect, resolve and replay them.
Approval gates support timeouts from 4 to 72 hours. The run is checkpointed while it waits, so nothing is held in memory.
Book a working session with our team. Bring a real use case — we'll build it with you and show the governance and ROI views alongside.
