Automate

Workflows & an AI agent org you govern

Build automations on a drag-and-drop canvas, organize AI as an org chart of supervised agents, and decide — action by action — whether the system runs on its own, asks a human first, or just advises.

Visual Workflows Per-action autonomy

Everything in Automate runs over one execution engine and one record — whether it's a deterministic step, an AI step, or a whole org of supervised agents. You'll find it under the Automate workspace in the left rail.

3autonomy modes

1execution engine

∞process versions kept

Visual Workflows (Process Agents)

Workflows are a drag-and-drop canvas you build without code. In the agent org they're called Process Agents — a versioned, visual process that runs a defined job.

Triggers

Start a workflow on an event — lead.created, deal.created, a health-score alert, and more.
Steps & branches

Chain conditions, actions, AI steps and branches. Split into parallel paths and fan them back in.
Versioned

Editing a live workflow bumps its version; in-flight runs pin to their own — history stays intact.
Guardrails

Wrap any AI step in ordinary conditions and approvals. The AI proposes; the workflow governs.

Every step type runs on a single engine: triggers, conditions, actions, AI steps and branches. Because AI and deterministic steps share the same canvas, you can hold a model's output to the same rules as everything else.

Example: a lead-handling process

Here's a Process Agent triggered by lead.created — qualify the lead, gate on its score, assign it, notify the owner, then enforce an SLA and escalate if it's missed.

flowchart LR
  T([lead.created]) --> Q[Qualify]
  Q --> G{Score gate}
  G -->|passes| A[Assign owner]
  G -->|below threshold| N1[Nurture]
  A --> N[Notify owner]
  N --> S{SLA met?}
  S -->|yes| D([Done])
  S -->|no| E[Escalate]
  classDef n fill:#ffffff,stroke:#2563eb,stroke-width:1.5px,color:#0B1024;
  class T,Q,G,A,N1,N,S,D,E n;

Owner-flagged, account-tunable

A process owner can flag specific step fields — like the score threshold, the SLA, or who gets notified — as overridable. See per-account overrides below.

The agent org

Muntri organizes AI as an org chart, not a pile of disconnected bots. Master Agents supervise; Process Agents run the work; Customer and User agents hold context.

flowchart TD
  M[Master Agent<br/>supervisor]
  M --> P1[Process Agent<br/>Leads]
  M --> P2[Process Agent<br/>Deals]
  M --> P3[Process Agent<br/>Support]
  M --> CA[Customer Agents<br/>one per account]
  M --> UA[User Agents<br/>one per user]
  classDef master fill:#ecfdf5,stroke:#059669,stroke-width:1.5px,color:#0B1024;
  classDef n fill:#ffffff,stroke:#2563eb,stroke-width:1.5px,color:#0B1024;
  class M master;
  class P1,P2,P3,CA,UA n;

Layer	What it is
Master Agent	The supervisor. Manages a set of Process Agents and the context agents beneath them.
Process Agents	The versioned visual workflows above — each runs a defined process (leads, deals, support, CS…).
Customer Agents	One per account — the "Ask Muntri" agent that knows a specific account.
User Agents	One per user — the agent that knows a specific rep and their book of business.

The hierarchy map

You navigate the whole org as a hierarchy map: Master Agents expand to their managed Process Agents, each carrying badges for how many steps are customizable and how many accounts have customized them. Summary cards roll up your Customer and User agents. Click any process to drill straight into its canvas.

Building the agents themselves

For the full agent-building experience — the Agent Hub, marketplace templates, tools and guardrails — see the AI Agents section.

Per-action autonomy

This is how you stay in control. For each governable action a workflow can take, you choose the mode — tenant-wide, applied across every Process Agent.

Auto

Run the action automatically. Best for low-risk, high-volume work.
Require approval

Pause the run and notify the right person (human-in-the-loop). Approving re-runs that exact step and the workflow continues.
Advisory

Log what the agent would have done and skip it. Great for a trial run before you turn autonomy up.

Actions are grouped for easy governance — outbound (send email, enroll in a cadence), internal (send a notification, start a supervising agent), and record changes (create a task, and more). Set each one to the level of trust you're comfortable with, and dial it up as the automation proves itself.

flowchart LR
  ST[Action step reached] --> M{Tier?}
  M -->|auto| R[Run it]
  M -->|approve| H[Pause · notify · await approval] --> R
  M -->|advisory| L[Log recommendation · skip]
  classDef n fill:#ffffff,stroke:#2563eb,stroke-width:1.5px,color:#0B1024;
  class ST,M,R,H,L n;

Adopt autonomy gradually

Start new automations in advisory, watch what they'd do, then promote the steps you trust to approve or auto — one action at a time.

Per-account overrides

Global processes, local nuance. When a Process Agent owner marks a step field as overridable, an account owner can set a bounded per-account value for it — right from the account's Automation tab.

The account keeps the global process, but tunes only the flagged fields — a higher SLA for a strategic account, a different notify recipient, a stricter score threshold.
Overrides are bounded: only owner-flagged fields on owner-flagged steps can be changed, and only by the account owner or an admin.
At run time Muntri resolves the global step config, then applies the account's override on top.

⌘K command palette

Press ⌘K anywhere to open the command palette — search, navigate, and create or do things in one keystroke.

Create records

Spin up an account, contact, deal or lead without leaving the page.
Do actions

Build a report, create a segment, add a custom field, or kick off an import — straight from the palette.

Next: govern the models and budgets behind all this in the AI Agents section — bring your own model, route per feature, and cap spend.