bench get-app https://github.com/rogerboy38/raven_ai_agent

[![Listed on Frappe Gems](https://frappegems.com/api/method/frappe_gems.seo.badge?app=rogerboy38%2Fraven_ai_agent)](https://frappegems.com/gems/apps/rogerboy38/raven_ai_agent)

Raven AI Agent

Raymond-Lucy AI Agent for ERPNext with Raven Integration - Enhanced with OpenClaw-inspired architecture and an Agentic-Design-Patterns intelligence layer.

Current Status

Latest Update: May 2026 | Version: 2.2 Production Deployment: Active on https://erp.sysmayal2.cloud

Recent Deployments

Date	Changes
2026-05-01	Agentic Design Patterns intelligence layer (Reflection, Planner, Coordinator, Goal Loop, Fallback, RAG, Guardrails) wired into agent_v2 (PR #3)
2026-03-21	Pipeline diagnosis commands (@ai pipeline, @ai diagnose), Payment Agent, Manufacturing workflow
2026-03-20	Data Quality Scanner, Sample Request from Lead/Prospect/Opportunity/Quotation/SO
2026-03-19	Payment Entry creation and submission fixes, @ai payment routing
2026-03-09	Sales Invoice workflow fixes (modeofpayment field), command routing corrections
2026-03-09	Quality Management System (QMS) field bug fixes
2026-03-08	Raven User synchronization for mobile/web parity
2026-03-08	Phase 4 Advanced Analytics & Reporting module initiated

Features

Core Protocols

• Raymond Protocol: Anti-hallucination with verified ERPNext data
• Memento Protocol: Persistent memory storage across sessions
• Lucy Protocol: Context continuity with morning briefings
• Karpathy Protocol: Autonomy slider (Copilot → Command → Agent)
• Agentic Design Patterns Layer: 7 provider-agnostic patterns (Reflection, Planner, Coordinator, Goal Loop, Fallback, RAG, Guardrails) that boost the agent's reasoning, planning and safety — see Intelligence Layer

Multi-Provider LLM Support

Provider	Models	Features
OpenAI	gpt-4o, gpt-4o-mini, gpt-4-turbo	Default provider
DeepSeek	deepseek-chat, deepseek-reasoner	Cost-effective, reasoning mode
Claude	claude-3-5-sonnet, claude-3-opus	Strong analysis
MiniMax	abab6.5-chat, abab5.5-chat	Multilingual
Ollama	llama3.x, qwen, mistral, etc.	On-prem / offline

All five providers transparently work with the FallbackChain pattern — if the primary provider fails or returns empty, the chain falls through to the next one in your configured order.

Multi-Channel Gateway

• Raven (Primary): Native ERPNext chat integration with AI commands
• WhatsApp Business API: Full messaging + interactive buttons
• Telegram Bot: Messages, voice, inline keyboards
• Slack: Direct messages, app mentions, button actions
• Session Management: Cross-channel context preservation

Voice Integration

• ElevenLabs TTS: Text-to-speech responses
• Multiple Voices: Rachel, Drew, Bella, and more
• Auto Voice Detection: Respond with voice when appropriate

Skills Platform

• Browser Control: Web automation and data extraction
• Extensible Architecture: Easy to add new skills
• Intent Routing: Automatic skill matching

Quality Management System (QMS)

The QMS module provides comprehensive quality control capabilities through Raven AI commands:

Command	Description
@ai quality setup status	View QMS configuration and status
@ai quality create nc	Create Non-Conformance report
@ai quality create training	Create Training Program
@ai quality create audit	Create Internal Audit

Verified Working (March 2026): - ✅ Non-Conformance Creation: QA-NC-00020 - ✅ Internal Audit Creation: QA-MEET-26-03-08 - ✅ Training Program Creation: GMP Basics

Pipeline & Diagnosis Commands

Full sales pipeline management from Quotation to Delivery:

Command	Description
@ai pipeline SAL-QTN-XXXX	Full pipeline diagnosis with status
@ai diagnose SAL-QTN-XXXX	Detailed diagnosis with issues and next steps
@ai check data SAL-QTN-XXXX	Validate quotation data quality
@ai fix SAL-QTN-XXXX	Auto-fix data quality issues
@ai scan SAL-QTN-XXXX	Full data quality scan
@ai validate SAL-QTN-XXXX	Validate data integrity
@ai repair SAL-QTN-XXXX	Auto-repair issues
@ai validate ACC-SINV-XXXXX	Validate sales invoice
@ai !fix SAL-QTN-XXXXX	Fix cancelled quotation
@ai !update quotation SAL-QTN-XXXX item ITEM-CODE	Update quotation item

Pipeline Stages: - Quotation → Sales Order → Work Order → Stock Entry → Delivery Note → Sales Invoice → Payment

Sales-to-Purchase Full Cycle

Complete sales pipeline from opportunity to payment and purchase requisition:

Sales Cycle

Command	Description
@ai show opportunities	List sales opportunities
@ai create opportunity for [customer]	Create new sales opportunity
@ai check inventory for [SO]	Check item availability for Sales Order
@ai show quotations	View your quotations
@ai show sales orders	View your sales orders
@ai show pending deliveries	Delivery notes, stock levels
@ai create delivery note for [SO]	Ship items to customer
@ai create sales invoice for [SO/DN]	Invoice the customer

Purchase Cycle

Command	Description
@ai create material request for [SO]	Create Material Request from SO
@ai show material requests	List pending material requests
@ai create rfq from [MR]	Create Request for Quotation
@ai show rfqs	List RFQs and their status
@ai show supplier quotations	List supplier quotations
@ai create po from [SQ]	Create Purchase Order from Supplier Quotation
@ai receive goods for [PO]	Create Purchase Receipt

Payment Management Agent

Complete payment workflow automation:

Command	Description
@ai payment create [SI-NAME]	Create Payment Entry from Sales Invoice
@ai payment create [SI-NAME] amount [AMOUNT]	Partial payment
@ai payment submit [PE-NAME]	Submit Payment Entry
@ai payment reconcile [PE-NAME]	Check reconciliation status
@ai payment outstanding	List all unpaid invoices
@ai payment outstanding customer [NAME]	Unpaid for specific customer
@ai payment status [PE-NAME]	Payment Entry details
@ai create payment for ACC-SINV-XXXX	Create Payment Entry from Sales Invoice
@ai validate ACC-SINV-XXXX	Validate sales invoice

Full Payment Cycle:

@ai payment create ACC-SINV-2026-00001
@ai payment submit ACC-PAY-2026-00001
@ai payment reconcile ACC-PAY-2026-00001

Manufacturing Workflow Agent

Automated manufacturing from Sales Order:

Command	Description
@ai work order from SO-XXXXX	Create Work Order from Sales Order
@ai transfer materials	Transfer raw materials to WIP
@ai manufacture MFG-WO-XXXXX	Complete manufacturing
@ai submit wo MFG-WO-XXXXX	Submit Work Order
@ai !submit Work Order MFG-WO-XXXX	Submit Work Order (direct)
@ai !submit bom BOM-XXXX	Submit Bill of Materials
@ai unlink sales order from MFG-WO-XXXX	Remove SO link from Work Order
@ai !cancel bom BOM-XXXX	Cancel submitted BOM
@ai !revert bom BOM-XXXX to draft	Reset cancelled BOM to draft

Sample Request Management

Create Sample Requests from any source document via button or command:

Command	Description
Create → Sample Request button	Lead, Prospect, Opportunity, Quotation, Sales Order
@ai sample request Lead LEAD-NAME	Create sample request from Lead
@ai sample request Prospect PROSPECT-NAME	Create sample request from Prospect
@ai sample request Opportunity OPP-NAME	Create sample request from Opportunity
@ai sample request Quotation SAL-QTN-XXXX	Create sample request from Quotation
@ai sample request Sales Order SO-XXXX	Create sample request from Sales Order

Features: - Auto-populates party, contact, address - Default item selection based on source type - Request type mapping: Marketing, Prospect, Pre-sample Approved, Representative Sample, Exhibition

Data Quality Scanner

Pre-flight validation and repair:

Command	Description
@ai scan SAL-QTN-XXXX	Full data quality scan
@ai validate SAL-QTN-XXXX	Validate data integrity
@ai repair SAL-QTN-XXXX	Auto-repair issues

Cost Monitoring

• Usage Tracking: Per-user token consumption
• Budget Alerts: Warnings when approaching limits

Comparison: ravenaiagent vs OpenClaw

Feature	ravenaiagent	OpenClaw
Integration	Frappe/Raven + Multi-channel	Multi-channel only
Architecture	Gateway + Session Management	Gateway/WebSocket
AI Backend	OpenAI + DeepSeek + Claude + MiniMax	Claude + OpenAI + local
Channels	Raven, WhatsApp, Telegram, Slack	WhatsApp, Telegram, Slack
Voice	ElevenLabs TTS	ElevenLabs
Skills	Browser control, extensible	Browser, canvas, device
Cost Monitor	✅ Built-in	❌
ERPNext Native	✅	❌

Installation

bench get-app https://github.com/your-repo/raven_ai_agent
bench --site your-site install-app raven_ai_agent

Configuration

AI Agent Settings

1. Go to AI Agent Settings in ERPNext
2. Select Default Provider and enter API keys
3. Set Fallback Provider for automatic failover
4. Configure Cost Budget for usage warnings

Channel Configuration (Optional)

# WhatsApp
whatsapp_config = {
    "phone_number_id": "YOUR_ID",
    "access_token": "YOUR_TOKEN",
    "verify_token": "YOUR_VERIFY_TOKEN"
}

# Telegram
telegram_config = {"bot_token": "YOUR_BOT_TOKEN"}

# Slack  
slack_config = {
    "bot_token": "xoxb-YOUR-TOKEN",
    "signing_secret": "YOUR_SECRET"
}

Voice Configuration (Optional)

voice_config = {
    "elevenlabs_api_key": "YOUR_KEY",
    "default_voice": "rachel"
}

Usage

In Raven

@ai What are my pending sales invoices?
@ai Show me top customers by revenue
@ai quality setup status
@ai !quality create training GMP Basics

Via API

from raven_ai_agent.api.agent_v2 import process_message_v2

result = process_message_v2("What invoices are due?")
result = process_message_v2("Analyze data", provider="claude")

Multi-Channel

from raven_ai_agent.channels import get_channel_adapter
from raven_ai_agent.gateway import session_manager

adapter = get_channel_adapter("whatsapp", config)
incoming = adapter.parse_webhook(payload)
session = session_manager.get_or_create_session(user_id, "whatsapp", incoming.channel_user_id)

Voice

from raven_ai_agent.voice import ElevenLabsVoice

tts = ElevenLabsVoice(api_key="YOUR_KEY")
audio = tts.text_to_speech("Hello!")

Autonomy Levels

Level	Name	Description
1	Copilot	Read-only queries, suggestions
2	Command	Execute with confirmation (use ! prefix)
3	Agent	Multi-step autonomous workflows

Important: Commands with ! prefix execute directly without confirmation. Always use @ai !command format in Raven channels.

Architecture

``` ravenaiagent/ ├── api/ │ ├── agent.py # V1 API (Raymond / Lucy / Memento) │ ├── agentv2.py # V2 API (multi-provider + intelligence layer) │ ├── workflows.py # Business workflow automation │ ├── commandrouter.py # Command routing logic │ ├── multiagentrouter.py # Regex pipelines + Coordinator semantic fallback │ ├── intentresolver.py # NL → command │ └── memorymanager.py # Persistent memory + vector search ├── patterns/ # Agentic Design Patterns intelligence layer │ ├── reflection.py # Producer / critic loop (Ch. 4) │ ├── planner.py # JSON pla