Frappe Flow: 7 Practical Ways to Put an AI Assistant to Work Inside ERPNext

Frappe has released Flow, an app that puts an AI assistant inside your Frappe and ERPNext site. Instead of clicking through lists and forms, you press Ctrl+I (Cmd+I on a Mac) and ask in plain English. The assistant finds the right records, reads them and, with your approval, creates and updates documents.

For a small or mid-size business, the value is not the novelty. It is a way to cut the routine work that fills a manager’s day. Here is what Flow is, seven practical ways to use it, and what to think about before you switch it on.

What Flow actually is

Flow has two parts. The first is the assistant: a chat panel in the Desk, driven by a built-in agent, that works within each user’s own permissions. The second is a framework of agents, tools, triggers, knowledge bases and memory that you set up from the Desk or in Python, so an assistant can work on a schedule or react to new documents without anyone asking.

It works with the major AI providers, and with local models through tools such as Ollama when your data must stay on your own servers.

7 practical ways to use it

1. Ask questions about your own data. “Show sales orders from last month that are not delivered yet.”
2. Do routine data work. “Create a follow-up ToDo for every quotation older than seven days that has no order.”
3. Clean up master data. “List items in the Raw Material group with no HSN code.” Review the list, then fix them in one approved step.
4. Put reporting on a schedule. Every weekday at 9 am, email the sales team yesterday’s new leads with contact details and source.
5. React to new documents. When a support issue arrives, classify it and suggest a first reply for the team to check.
6. Answer from your own documents. Index product manuals, policies or resolved tickets so an agent answers from your material, not from guesswork.
7. Configure the system in plain English. “Add a Region field to Customer.” Flow creates the custom field instead of you hunting through settings.

How it stays safe

• Permissions: the assistant acts with the user’s own permissions. If you cannot see or change a record, neither can the assistant working for you.
• Approvals: any change, such as creating, updating or deleting records, pauses and asks you to approve it first.
• Audit: every conversation and every step is saved, including what the assistant did and how many tokens it used.

Agents that run on their own, on a schedule or on new documents, are the one place to be careful. Because nobody is there to approve their actions, give them a dedicated user with only the permissions they need, and test them on a copy of your site first.

What to think about before you switch it on

• Data privacy. What the assistant reads is sent to your AI provider, unless you run a local model. Decide what is acceptable for your business.
• Cost. Every run records its token usage. Use smaller models for routine scheduled jobs and stronger ones for complex work.
• Start small. Pick two or three clear use cases and prove them before rolling Flow out to everyone.
• Review the runs. Look at what the assistant did each week while you are learning. It is the fastest way to improve your instructions.

Getting started

Flow runs on Frappe version 16 and later. You install the app on your bench, connect an AI provider or a local model, and the built-in assistant appears in the Desk. From there, the real work is choosing good use cases and setting them up safely.