erpnext_copilot

KO | EN

gitlite — search

#python #java #python3 #arduino #golang #machine-learning #rust #html #flask #javascript #seismology #nodejs

erpnext_copilot

★ 11 Open GitHub ↗

AI-powered ERPNext assistant — query ERP data, perform actions, generate charts, and automate reports using natural language.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ERPNext Copilot

Ask ERPNext in plain English, safely.

## Features

- Desk page to ask natural-language questions about your ERP data
- Schema-aware planning (real DocType metadata, not guessed fields)
- Structured JSON query plans validated before execution
- Permission-safe reads via `frappe.get_list` / `frappe.get_doc` as the **current user**
- OpenAI-compatible providers (OpenAI, Azure, Groq, Ollama, …)
- Demo prompts for quick screenshots / community demos
- Optional draft **create** with explicit Approve / Cancel

## 2-minute installation

```bash
cd frappe-bench

# Folder name must match the app: erpnext_copilot
git clone https://github.com/ameenmari/erpnext_copilot.git apps/erpnext_copilot

bench setup requirements
bench --site <site> install-app erpnext_copilot
bench --site <site> migrate
bench build --app erpnext_copilot
bench restart
```

Requires a working Frappe bench with **ERPNext** installed for invoice/stock demo questions.

## Configure AI

1. Open **ERPNext Copilot Settings** (or Workspace → ERPNext Copilot → Settings)
2. Check **Enabled**
3. Set **Provider** = OpenAI (OpenAI-compatible)
4. Set **Model** (e.g. `gpt-4o-mini`)
5. Paste **API Key**
6. Optional **Base URL** for compatible endpoints (`http://127.0.0.1:11434/v1` for Ollama)
7. Save

Assign role **ERPNext Copilot User** to people who should ask questions (Administrator gets it on install).

## Ask your first question

Desk → **ERPNext Copilot** → type a question → **Ask**.

## Example questions

- Show unpaid invoices older than 30 days
- Show unpaid invoices
- Show low-stock items
- Show this month's sales invoices
- How many sales orders were created this week?
- Show customers with overdue payments
- Show purchase orders waiting for approval
- Show today's deliveries
- Find customer ABC

## How safety works

AI proposes a structured query → validator checks it → ERPNext executes it using the current user's permissions.

- No raw SQL by default (not implemented)
- No generated Python execution
- No `ignore_permissions`
- Mutations (create) require on-screen **Approve**

## Privacy

Sent to the LLM by default:

- your question
- relevant DocType field metadata
- today's date / capabilities

ERP business **records** are queried locally inside ERPNext and are not sent to the model for planning.

## Current limitations

- Read-focused MVP (`list`, `count`, `get`); create is draft + approval only
- No delete / cancel / submit / payment / refund
- Stock “low stock” quality depends on Item fields available on your site
- Requires an external LLM (or local OpenAI-compatible server)
- Best with ERPNext DocTypes installed (Sales Invoice, Item, …)

## Development

```bash
# Pure unit tests (no bench required)
cd apps/erpnext_copilot   # or this repo root
python -m unittest erpnext_copilot.tests.test_json_plan -v

# Full app tests (Frappe site required)
bench --site <site> run-tests --app erpnext_copilot
```

## License

MIT — see [license.txt](license.txt).

## Security

See [docs/SECURITY.md](docs/SECURITY.md).

erpnext_copilot

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README