Visual Workflow Studio
Visual Workflow Studio is where you design, run and schedule governed agentic workflows. You compose the steps an agent takes, and every action step remains governed by Gateway. It turns the zero-trust model into something you can build, review, run, watch and stop, before and while it executes.
The Studio opens as its own page (the Workflow Studio link in the app top bar). It needs an editable, Advanced-tier session with Agentic CISO access, and the platform must be running in its database mode.
The building blocks
Section titled “The building blocks”A workflow is assembled from a small set of governed objects, each held in Agentic CISO and each carrying its own policy:
| Object | What it is |
|---|---|
| Workflow plan | The overall design: a name, objective, root agent, and the ordered steps. |
| Workflow step | A single step with a task type. Action steps reference a governed connector by tool name. |
| Task type | What a step does, for example governed reasoning, tool plan, evidence collection, investigation, or a single Gateway tool call. |
| Toolset / connector | A governed connector adapter with allowed actions and data classes, Gateway-enforced, with approval required for high-risk actions. |
| Checkpoint | A saved, resumable state with a rollback policy, so long-running work can pause and resume inside the trust boundary. |
The Plan Runtime Wizard
Section titled “The Plan Runtime Wizard”The fastest way to a working runtime is the Plan Runtime Wizard, a guided panel at the top of the Studio. It steps you through eight stages and, at each gate, tells you exactly what to fix, with a button that takes you straight to where the fix is made.
- Prerequisites. Confirms the runtime is ready: Claw and Gateway configured and reachable, tool audit logging enabled, and a connector in proper credential custody. Each blocker shows a Fix it button (for example Open Tool Permissions) and a warning does not block you.
- Starting point. Choose Bootstrap the governed demo (provisions a working agent, its tool permissions and a plan that already passes preflight, the fastest path to a first success), start from a template, or start blank.
- Define plan. Set the plan name, objective, risk level, root agent (the orchestration identity), maximum parallel steps and timeout.
- Build steps. Add steps from the task-type palette, set each step’s agent, tool (for a Gateway tool call), dependencies and whether it needs approval. A wave preview shows the execution order and flags any dependency cycles.
- Preflight and fix. Saving the plan runs the authoritative server preflight, which checks every step against exactly what Gateway will enforce at runtime. Anything missing is listed as a categorised checklist, each item with a clickable fix that opens the right Agentic CISO screen (with the specific agent pre-selected for per-step issues).
- Approve. The governance sign-off, recorded in the audit trail. It attests the plan passed preflight under zero-trust enforcement.
- Dispatch and run. Sends ready steps to the runtime. See Running a plan.
- Done. A summary of the run, with the option to create another.
Every gate is enforced by the server, not just the wizard: you cannot reach a confirmed, dispatched runtime without passing the real preflight and approval.
Setting up and running a plan
Section titled “Setting up and running a plan”Prerequisites
Section titled “Prerequisites”Before a plan can run, an operator and an administrator make sure the runtime is in place. The wizard’s first step verifies all of this for you:
- The Claw execution service and the Gateway enforcement service are configured and reachable (an operations task, see Claw and Gateway).
- At least one connector is configured, with its credential held in Gateway custody (configure these in the Studio’s connector catalogue panel).
- Every registered agent that a step uses has a named human owner, an assigned ATF level, and the required Tool Permissions active with audit logging enabled.
- Material action steps have matching Runtime Access Policy coverage and the agent is mapped into the Org Chart where high-impact escalation is required.
- For any step that needs human sign-off, an approval gate exists.
If something is missing, the wizard’s checklist links you straight to the screen that fixes it.
Build the plan
Section titled “Build the plan”- Open Workflow Studio and select your workspace.
- Use the wizard’s Bootstrap demo for a guided first run, or pick a template, or build from scratch.
- In Define plan, choose the root agent and set parallelism and timeout.
- In Build steps, add each step: choose its task type, assign a registered agent, set a tool for Gateway tool calls, mark dependencies (which steps it runs after), and tick requires approval for sensitive steps.
Preflight and fix
Section titled “Preflight and fix”Save the plan to run preflight. The readiness scorecard shows a score and which steps are ready. For anything not ready, the checklist tells you the precise problem and gives a Fix it button:
| Blocker | Where the fix-it button takes you |
|---|---|
| Tool permission not active / not runtime-enabled / audit logging off | Tool Permissions (Agentic CISO), with that agent focused |
| Runtime Access Policy coverage missing | Access Matrix (Agentic CISO), with that agent focused |
| Org escalation path missing | Org Chart (Agentic CISO), with that agent focused |
| Agent or root agent not registered | Agent Register |
| Approval gate missing | Approval Gates |
| ATF level | ATF assessment |
| Connector missing | the Studio’s connector catalogue |
| Draft issues (step keys, dependencies) | the wizard’s Build step |
Fix, return, and re-check. The plan is ready when preflight passes.
Approve and dispatch
Section titled “Approve and dispatch”Approve the plan (the governance sign-off), then dispatch. Dispatch sends each ready step to the runtime under a signed runtime identity, and Gateway re-validates every action at dispatch, so a step can still be blocked or held for approval at runtime.
Running a plan
Section titled “Running a plan”Steps run in waves by dependency: steps with no unmet dependencies run first, then the next wave once they complete.
- Auto-run (default on): the Studio automatically dispatches each next wave as the previous one completes, so the whole plan runs to completion on its own. Turn it off to step through manually, one wave at a time, with Sync + continue.
- Approval-gated steps still wait for a human on every run, and every Gateway decision still holds, so Auto-run never bypasses governance.
- A live stream shows each step’s Gateway decisions, model calls and bounded output, backed by a hash-chained Claw journal.
Stopping a running step
Section titled “Stopping a running step”A running step is backed by a Claw task. Each active step has a Stop button that cooperatively cancels it: Claw stops at the next safe checkpoint, marks the task cancelled and will not half-complete. The step is terminal, but you can re-dispatch the plan afterwards. Stopping is governed and audited.
For a broader halt, an administrator can use the Gateway kill switch (admin console, see Gateway), or suspend an individual agent in Agentic CISO.
Scheduling a plan
Section titled “Scheduling a plan”An approved plan can run on a recurring schedule. On the selected plan’s Schedule panel:
- Choose a cadence (hourly, daily, weekly, monthly or quarterly).
- Click Enable schedule. Enabling is a sign-off and needs an approver-level permission plus an approved plan.
- Pause, Resume or Disable the schedule at any time. The panel shows the next run and the last run with its outcome.
A schedule is never a standing bypass. On every scheduled run, Gamut:
- re-checks the schedule owner’s live permission and the tenant entitlement (and pauses the schedule if either was revoked),
- re-runs the full preflight,
- skips the run if a previous run is still in flight (no overlap),
- skips while the kill switch is active (runs resume automatically once it is cleared),
- and dispatches through the same Gateway/Claw-enforced path, where per-step approvals still hold.
Every scheduled run is audited with its outcome.
Deleting a plan
Section titled “Deleting a plan”Each saved plan in the sidebar has a delete control. Deleting removes the plan and its steps (with a confirmation) and is audited. Any Claw tasks already running are governed independently and are unaffected.
Governed by construction
Section titled “Governed by construction”A workflow built in Studio does not bypass the agentic stack. Each action step references a governed connector by tool name, and that call faces the same checks as any other: the tool must be permitted to the agent, Gateway must allow the action within the agent’s ATF level boundary, a Runtime Access Policy must cover the action, the org chart must support high-impact escalation, and any approval gates must be satisfied. The design surface changes; the enforcement model does not.
Runtime backends
Section titled “Runtime backends”A workflow targets a registered runtime backend, either Gamut Claw or, for external frameworks, a BYO runtime. Backends are governed: they default to a deny network posture, no raw secret mounts, and gateway-only egress. In every case the rule holds: think anywhere, act through Gateway.
- Agentic CISO: register the agents, tools and approval gates a plan uses.
- Connector catalog: the tools workflow steps can call.
- Gamut Claw and Gateway: the runtime and enforcement backends.