Agent Workflow Studio overview

Build governed, tenant-scoped agent workflows that run on schedules, business events, or manual input, with human approval for external writes.

Agent Workflow Studio overview

The Agent Workflow Studio is in a staged rollout. In production, the Agents entry in the main navigation and the /agents route are currently visible only to internal Control staff. If you don’t see Agents in your sidebar yet, this is expected — talk to your Control contact if you’d like to know when it opens up for your tenant.

The Agent Workflow Studio is where you create, publish, and monitor Control’s agent workflows. Each workflow is a tenant-owned definition that reasons over Control data, runs on schedules or business events, and can prepare Slack or Gmail actions for a reviewer to approve before anything leaves Control.

Open the Studio from the Agents entry in the main navigation.

When to use an agent workflow

Use an agent workflow when the same finance task happens on a schedule or business event and its output can be described as instructions plus a small set of Control data reads. Good fits include:

• Building a monthly management summary as soon as the month is ready to close.
• Sending a daily digest of KPI movements and unusual counterparties.
• Reviewing new counterparties against Control’s transaction history.
• Reacting to a signed external event from another system.

A workflow is not the right home for one-off ad hoc questions. Use AI chat and canvas for those.

What the Studio contains

The Studio has five views, available as tabs at the top of the page:

View	What it shows
Agents	Every workflow in the tenant, its status, current version, and next scheduled run. Start new workflows from templates or blank drafts.
Builder	The draft editor for the selected workflow: name, instructions, output contract, data tools, triggers, and validation.
Approvals	External-action proposals waiting for a human decision, plus recently executed or rejected proposals.
Runs	Persisted run history for the selected workflow, including start and completion timestamps, events, and error details.
Connections	Slack, Gmail, and other integrations linked to the tenant through Nango, plus signed event credentials for external triggers.

The header shows the organization-wide AI processing policy currently in effect for agents. Open to change it. AI must be explicitly enabled for the organization before workflows can make AI-processing requests.

How a workflow moves through its lifecycle

Template or blank
  → Draft (edit instructions, tools, triggers)
  → Validate
  → Publish immutable version
  → Trigger (manual, cron, event, or signed external event)
  → Occurrence
  → Run (with events, external-action proposals, and audit entries)

Key properties of this lifecycle:

• Drafts are mutable. You can rename, re-instruct, and re-scope tools until you publish.
• Published versions are immutable. Publishing seals the instructions, output contract, tool grants, and trigger set into a numbered version with a snapshot checksum. Later edits create a new draft that becomes the next version when published.
• Triggers produce deduplicated occurrences. The same cron tick, month-close signal, or signed external event will not run the workflow twice.
• Every run is persisted. You can always see what started a run, what it fetched, what it produced, and where it failed.

Governance model

Agent workflows never gain ambient write access to Control or connected systems.

• Data reads are limited to the tools you explicitly enable in the builder. Tools are marked with a risk label so you can see which reads are sensitive.
• External writes (for example, posting to Slack) never execute directly. The agent files a proposal in the Approvals view with the exact payload, a proposal hash, and a fixed destination. A human approves or rejects it, and Control executes the approved payload atomically so it cannot run twice.
• Tenant isolation applies everywhere. Workflows, connections, event credentials, occurrences, and runs are scoped to the tenant, and cross-tenant reads are blocked.
• AI processing location follows the tenant’s organization-wide policy. The active policy is shown in the Studio header.
• Audit trail. Every run records the events it emitted, the tools it invoked, the proposals it filed, and the decisions taken on them.

Availability

Attribute	Value
Location	Agents in the main navigation.
Roles	Any tenant member can open the Studio and manage workflows. Creating and revoking signed event credentials requires an organization admin.

Next steps

• Build and publish a workflow — go from template to a published version.
• Trigger workflows — configure manual, cron, month-close, and signed external triggers.
• Review runs and approvals — inspect run history and decide on external-action proposals.
• Connect Slack, Gmail, and event sources — link accounts through Nango and mint signed event credentials.