Where Activities run and Tasks land.
The runtime where the work actually happens. Activities show their live state, step by step. Tasks queue up in front of the right person with the document and system context attached. Exceptions get reviewed by your team or worked by an agent. One audit trail across all of it.
Live state on the right. Tasks on the left. Audit trail underneath.
Workflow is two views in one screen. The left rail is the queue: Tasks waiting for a person, sorted by who owns them and what they're for. The right pane is the running Activity: the steps, what each step did, what's still in flight. Click a Task and you see the document it relates to, the rules that fired, and the prior steps that led to it.
Activity: Northwind Foods · 2026-Q3 covenant pack
Running doc-NWF-Q3-26-pack.pdf · 38 pagesThe same screen during an extraction run, an analyst review, and an audit-trail lookup.
The right work, in front of the right person, with the context attached.
A Task isn't a notification. It's a row a person owns, with the document open, the prior steps visible, the rules that fired in view, and the action the platform's asking for already framed. The analyst's job is the judgment call. The platform brings everything else.
- Tasks queue per-person, with ownership rules per process.
- Document and system context attach to the Task automatically — no clicking out to fetch the file.
- Status changes are first-class. A Task moving from "In Review" to "Approved" can fire the next Activity.
- Reassignment, comments, and history sit on the Task itself.
Every step. Every input. Every result.
An Activity is a typed graph of steps with declared inputs and outputs. Workflow shows it live: which step ran when, what it produced, where the run is now, and where it stopped if it stopped. Re-run a step in isolation when you need to.
- Step kinds: extraction, validation, scripted, bridge call, LLM, agent, approval, create-Task.
- Run state visible per step: not started, running, succeeded, failed, waiting on Task.
- Inputs and outputs inspectable at every step.
- Re-run from any step without restarting the whole Activity.
Exceptions that are routine come off the queue.
When an exception fires, an agent has the first look. If the case matches an existing rule in Knowledge or matches a pattern your team has cleared before, the agent works it and notes the action on the audit trail. If it's a judgment call, the agent escalates to a Task. The result is a queue that surfaces what your team should actually be looking at.
- Agents act inside the same Activity Plan that humans run; the Plan governs them.
- Every agent action is logged to the audit trail with the rule or pattern it matched.
- Your team sets the threshold for what an agent can clear without review.
Automated and human work. Same trail.
Every step in an Activity, every Task action, every agent decision, every Knowledge rule that fired — recorded once, in the order it happened, against the document and the workflow run. Pull the audit for a single document or for a quarter's worth of work; it's the same query.
The platform can't do anything your team couldn't do. Same rules, same guardrails, same audit trail.
See Workflow on your process.
We'll run a real Activity Plan against documents that look like yours and walk you through the queue, the run state, and the audit trail.