Built for the
Last Mile.
Record browser workflows. AI understands what you're doing. Run it again — perfectly.
For finance and back-office teams inside GCCs, BPOs, and enterprises — stuck behind the CoE backlog queue.
Most manual back-office work isn't hard to automate.
It just never paid off.
India's commerce graduates and back-office operations staff spend hours every week on the same structured, repetitive work — GSTR filings, bank reconciliations, accrual close checklists, vendor statement matching. The same steps, every cycle. Predictable. Automatable. Yet unautomated.
The CoE backlog is 6–12 months long. Even when a request is picked up, delivery takes another 3–4 months. Traditional RPA needs ~500+ hours/year saved to justify its ₹1–5 lakh implementation cost. Most last-mile work never clears that bar. So it stays manual. Forever.
20–25%
Already automated
High-volume, high-frequency back-office. Enterprise RPA — UiPath, Automation Anywhere, Blue Prism — already handles this. Justified ROI, developer-built, CoE-owned.
50–55%
The Last Mile
Low-volume, fully structured, fully repetitive. Never automated — traditional RPA can't justify the cost, CoE never has bandwidth. This is exactly where Guerrilla Bots lives.
20–25%
Judgment work
Genuinely unstructured. Needs real human or AI judgment to handle exceptions, grey areas, and novel decisions. Stays with people and AI agents.
The Cursor for RPA.
Most of the automation builds itself. A developer only finishes the last mile — the same relationship Cursor has with a developer.
Record
Click through the task once. I² Recorder captures every browser action — click, type, navigate, wait — and labels each step with AI. One recording session becomes a portable workflow file.
Run
Operative replays the workflow locally on your machine. Credentials stay in a local vault. Runs on your existing Windows machine and IT-approved Python setup. Nothing touches the cloud during execution.
Recover
If a step breaks, AI retries first. If that fails, Deputy pauses the run, a human steps in and fixes just that step, then the bot resumes from exactly where it stopped. Every run finishes.
Five tools. One workflow.
I² Recorder
Record any browser workflow — no code.
Click through your task once. I² Recorder captures every action, labels each step with AI, and packages it into a portable workflow file (.gbw). Works on any web application — GST portal, HRMS, bank portals, ERP front-ends. The strongest proof that automation doesn't need a developer to start.
Operative
The local agent that runs your workflows.
A desktop tray app that replays recorded workflows on your machine. Credentials stay in a local vault — never in the cloud. Runs on infrastructure IT already approves: Python on your existing Windows machine. No new server, no new risk surface.
Workbench
Edit workflows without re-recording.
Simple mode for business users — swap a selector, change a value, reorder steps. Advanced mode for developers — add branching logic, variables, validation rules, conditional paths. One editor, two levels of depth. The developer only touches the 10–20% that needs real logic.
Debrief
Extract the logic already inside your Excel files.
Most process knowledge is buried in Excel. Debrief reads the workbook behind a process and surfaces the business rules embedded in formulas, conditional formatting, and named ranges — ready to wire into a workflow. Understand the process before you automate it.
Deputy
Nothing silently breaks.
When a bot fails, AI retries first. If that fails too, the run pauses and a human steps in — fixes the one broken step — then the bot resumes from exactly where it stopped. Full audit trail on every exception. The difference between an automation that runs and one that finishes.
Vigil
The governance layer for your CoE.
Exception inbox, full run history, and audit trail in one place. The layer that lets a CoE formally adopt Guerrilla Bots instead of treating it as shadow IT. Every run logged, every exception tracked, every fix recorded.
Let's talk.
We scope the right fit together — number of workflows, users, and deployment model. No number on a webpage is more useful than a 20-minute conversation.
Questions we actually get.
Is this shadow IT?
+Do I need to know how to code?
+What happens when the automation breaks?
+How is this different from UiPath or Power Automate?
+How long does it take to automate something?
+Where does my data go?
+What if our process changes?
+