QubitPilotBeta
ProductFor CAsFor advocatesPricingDocs
Practice essentials
  • Clients
  • Cases
  • Tasks
  • Calendar
  • Billing
Court Intelligence
  • Court Intelligence
AI & automation
  • Assistant
  • Canvas
  • Knowledge
  • Automations
  • Approvals
  • AI Tools
Files & settings
  • File Manager
  • Settings
Team & access
  • Roles & permissions
Home/Docs/Automations
AI & automation

Automations

Automations are AI agent workflows you either build from scratch or start from a template, then run once or put on a schedule.

Last checked against the product 31 August 2026 by Harsh Kalia

On this page
  • Starting from scratch or a template
  • Approval policy and step limits
  • Running an automation
  • When a run needs your approval
  • How to
  • Good to know

Starting from scratch or a template

An automation is defined by a name and a goal written in plain language. The goal is what the agent is trying to accomplish, not a fixed script. You can write a goal from scratch, or start from one of six built-in templates: weekly case summary, overdue invoice follow-up, new client onboarding, hearing preparation, daily task digest, and time entry audit.

Every automation also has a trigger: either manual (you run it yourself when you want it) or scheduled. A scheduled automation runs at a time you set, either every day or on specific weekdays.

Approval policy and step limits

Each automation carries its own approval policy, independent of any other automation: auto (the agent proceeds without pausing), review writes (the agent pauses before any action that changes data), or review all (the agent pauses before every step, including reads). This policy is what decides whether a run stops mid-way and waits for you.

A max steps setting, from 1 to 50 and defaulting to 16, caps how many actions a single run can take before it stops on its own. This bounds a runaway or looping goal.

  • Trigger: manual or scheduled (daily or specific weekdays, with a time)
  • Approval policy: auto, review writes, or review all
  • Max steps: 1–50, default 16

Running an automation

A saved automation runs with Run now, or Dry run. A dry run still executes the agent's full reasoning and produces a plan, but is explicitly labeled as not carrying out the side-effecting actions in that plan. It's a way to see what the automation would do before it actually does it.

You don't have to save an automation before running it. Quick run lets you type a goal and run it immediately as a one-off; afterward you're offered the option to save it as a proper automation, pre-filled with a title based on the goal you ran.

When a run needs your approval

If an automation's approval policy calls for it, a run pauses mid-way on a step and waits for you to approve or reject it, using the same mechanism as the Approvals inbox. You can act on it right from the automation's own run view, or from Approvals.

Approvals raised by an automation are handled slightly differently from approvals raised in a chat: an automation-sourced approval can never create a chat 'always allow' rule. That workspace-wide shortcut is chat-only. Every automation run always asks again, no matter how many times you've approved the same kind of step before.

How to

How to create an automation from a template

  1. Open Automations and choose one of the six templates (for example, 'Overdue invoice follow-up').
  2. Review and adjust the pre-filled goal, then set the trigger, approval policy, and max steps.
  3. Save the automation. It now appears in your list, ready to run manually or wait for its schedule.

How to try a goal without saving it first

  1. Use Quick run and type what you want the agent to do.
  2. Watch it run: a dry run shows the plan without acting, a live run carries it out subject to the approval policy.
  3. If it did what you wanted, save it as an automation from the prompt that appears afterward.

Good to know

  • Enable/disable, duplicate, and delete are all available directly from the automations list. Disabling one stops its schedule without deleting its history.
  • Dry run still runs the agent's real reasoning end to end; only the side-effecting actions themselves are held back.

Related modules

  • ApprovalsThe inbox of every AI or automation action waiting on a human decision: approve or reject, nothing runs without you.
  • AI ToolsA read-only observatory of every agent run and every capability the AI can use: the audit trail, not where decisions happen.
  • AssistantThe full-page AI chat that can look up and change workspace records under a mode you control, with every write pausing for approval.
  • CanvasCollaborative rich-text notes with real-time multi-user editing and an inline AI autocomplete you can turn on while you write.
QubitPilotBeta

Practice on autopilot - clients, matters, hearings, billing and an AI associate in one calm workspace, built for Indian advocates and CAs.

Start free trialTalk to us

5th Floor, Gordhan Sky Mall, Office 507, Khatipura Rd, Jhotwara, Jaipur, Rajasthan 302012 · +91 8000243808

GSTIN 08COWPC8765E1ZC

Find us on Google · LinkedIn

Product

  • Features
  • Pricing
  • Business plan
  • Integrations
  • Sign in
  • Get started

Resources

  • Docs
  • Help center
  • Blog
  • FAQ
  • Editorial Policy

Company

  • About
  • Team
  • Security
  • Book a demo
  • Contact

Legal

  • Privacy Policy
  • Terms & Conditions
  • Cancellation & Refunds

© 2026 Qubitron Labs · QubitPilot. All rights reserved.

QubitPilot