AI Model Providers (China & Self-hosted) - 支持DeepSeek/Qwen/Kimi/GLM/Doubao等、本地大模型及自定义模型

AI Model Providers (China & Self-hosted) is a unified AI model provider layer for Odoo, designed for Chinese users, global users, and self-hosted LLM deployments. This module extends Odoo AI by providing a flexible and extensible provider architecture that supports mainstream Chinese AI models, local/self-hosted LLMs, custom APIs, and embedding models. It seamlessly integrates with all Odoo AI features such as chat, AI agents, embeddings, and RAG workflows.

Supported AI Providers

This module adds support for the following AI providers with official API support:

• DeepSeek - Advanced AI models for chat and coding tasks
  • Models: deepseek-chat, deepseek-coder, deepseek-reasoner
  • High performance and cost-effectiveness
• Qwen (通义千问) - Alibaba Cloud's powerful AI model
  • Models: qwen-turbo, qwen-plus, qwen-max, qwen-max-longcontext
  • Supports both compatible mode and native DashScope API
  • Excellent Chinese language support
• ERNIE (文心一言) - Baidu's comprehensive AI model
  • Models: ernie-bot-turbo, ernie-bot, ernie-bot-4, ernie-bot-8k
  • Strong natural language understanding
• GLM (智谱AI) - Zhipu AI's intelligent models
  • Models: glm-4, glm-4-flash, glm-3-turbo
  • Multi-modal capabilities
• Moonshot (月之暗面) - Moonshot AI's advanced models
  • Models: moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k
  • Long context support (up to 128K tokens)
• Doubao (豆包) - ByteDance's versatile AI model
  • Models: doubao-seed-1-8-251228, doubao-seed-1-6-flash-250828, doubao-seed-1-6-lite-251015, doubao-seed-1-6-251015, doubao-1-5-pro-32k-250115
  • Fast response times
• Local LLM (本地LLM) - Support for self-hosted LLMs (Ollama, LocalAI, etc.)
  • Models: llama2, llama3, mistral, mixtral, qwen, codellama, phi, gemma, and custom models
  • OpenAI-compatible API support
  • Optional API key (Ollama doesn't require authentication by default)
  • Custom model name support
  • Default base URL: http://localhost:11434/v1 (Ollama)
  • ⚠️ Important for RAG: For optimal RAG (Retrieval-Augmented Generation) performance, it is strongly recommended to use local models with 14B parameters or larger (e.g., qwen3:14b, llama3:70b, mistral:large). Smaller models (8B or less) may not effectively utilize RAG context and may provide incomplete or inaccurate answers based on document content.

Key Features

Installation

1. Install the module from Odoo Apps
2. Go to Settings > AI > Providers
3. Enable and configure your desired AI providers with API keys
4. Select your preferred LLM model in AI Agent settings

Configuration

After installation, configure your AI providers:

1. Navigate to Settings > AI > Providers
2. In the Cloud Service Providers section, for each LLM provider you want to use:
   • Enable the provider toggle
   • Enter your API key
   • (Optional) Set a custom API base URL if needed
3. In the Local LLM Configuration section, configure up to 3 local LLM models:
   • Enable each local model configuration
   • Select a predefined model name or choose "Custom Model" to enter a custom model name
   • Enter the Base URL (required, e.g., http://localhost:11434/v1 for Ollama)
   • Optionally enter an API key if your local LLM service requires authentication
   • The configuration table uses full-width layout without horizontal scrollbar for better usability
   • ⚠️ For RAG Usage: If you plan to use RAG (document-based Q&A), strongly recommend using models with 14B parameters or larger (e.g., qwen3:14b, llama3:70b). Smaller models (8B or less) may not effectively utilize RAG context.
4. In the Embedding Providers (RAG使用) section, configure embedding services independently (for RAG usage):
   • Enter the embedding model name (e.g., text-embedding-3-small, text-embedding-v2)
   • Optionally configure API key and base URL for the embedding service
   • If not configured, the corresponding LLM provider's settings will be used as fallback
   • This allows you to use different endpoints for embeddings and chat
   • Note: The embedding model name field will not show invalid default values (e.g., "api")
5. Save the settings
6. Only configured providers/models will appear in AI Agent LLM model selection

Provider-Specific Notes

DeepSeek

• Supports chat and coding tasks
• For file uploads and embeddings, requires OpenAI API key configuration (uses OpenAI embedding API as fallback)
• Pure chat scenarios work without OpenAI configuration

Qwen (通义千问)

• Supports both native DashScope API and OpenAI-compatible API
• Automatically detects API type based on base URL configuration
• Default: Compatible mode (https://dashscope.aliyuncs.com/compatible-mode/v1)
• Native API: Set custom base URL to https://dashscope.aliyuncs.com/api/v1
• Automatic message chunking for long inputs (30720 character limit)
• Full embedding support

ERNIE (文心一言)

• Uses Baidu's access token authentication
• Full embedding support

GLM, Moonshot, Doubao

• OpenAI-compatible API format
• Full embedding support
• Standard Bearer token authentication

Local LLM (Ollama, LocalAI, etc.)

• Supports self-hosted LLM servers with OpenAI-compatible API
• API key is optional (Ollama doesn't require authentication by default)
• Default base URL: http://localhost:11434/v1 (for Ollama)
• Supports predefined models (llama2, llama3, mistral, etc.) and custom model names
• Up to 3 local models can be configured simultaneously
• Can configure separate embedding base URL if embedding service uses different endpoint
• If embedding base URL is not configured, uses the LLM provider's base URL
• Optimized configuration table: Full-width layout without horizontal scrollbar, with flexible column widths for better readability
• Configuration persistence: Local LLM settings are automatically preserved across module updates with intelligent data management
• ⚠️ Critical RAG Performance Requirement: For optimal RAG (Retrieval-Augmented Generation) performance, it is strongly recommended to use local models with 14B parameters or larger (e.g., qwen3:14b, qwen3:32b, llama3:70b, mistral:large). Smaller models (8B or less, such as qwen3:8b) may not effectively utilize RAG context and may provide incomplete, inaccurate, or generic answers that ignore the document content. While smaller models may work for basic chat, they often fail to properly understand and use the RAG context for document-based questions.

Usage

Once configured, you can use any of the supported providers in:

• AI Chat conversations
• AI Agent configurations
• Document embedding and RAG
• All other Odoo AI features

How Embedding Works

When you upload files to an AI Agent or use RAG (Retrieval-Augmented Generation):

1. The system automatically detects which embedding model to use based on your configuration
2. It first checks if the provider's default embedding model is configured in Embedding Providers
3. If not found, it searches for any configured embedding model and automatically associates it
4. If no embedding is configured, it falls back to using the LLM provider's API key and base URL
5. If neither is configured, you'll receive a clear error message with instructions on how to configure it

Example: If you're using Qwen as your LLM provider and have configured "text-embedding-3-small" in Embedding Providers, the system will automatically use that embedding model when processing uploaded files.

Troubleshooting

No models available in AI Agent: Make sure you have configured at least one provider (API Key or Base URL) in Settings > AI > Providers. Only configured providers will appear in the model selection.

DeepSeek file upload issues: If you're using DeepSeek and need to upload files, make sure to configure OpenAI API key in Settings > AI > Providers. DeepSeek uses OpenAI's embedding API for file processing.

Local LLM connection issues: Ensure your local LLM server (e.g., Ollama) is running and accessible at the configured base URL. Check firewall settings if connecting from a different machine.

Local LLM not using RAG context properly: If your local LLM is not effectively using RAG context (ignoring document content, providing generic answers), this is likely due to using a model that is too small. Use models with 14B parameters or larger (e.g., qwen3:14b, llama3:70b) for optimal RAG performance. Smaller models (8B or less) often cannot effectively process and utilize RAG context, even when it is correctly provided.

Module text still in Chinese when switching to English: Make sure to update the module after installation to load the English translations. Go to Settings > Apps, find the module, and click "Upgrade".

Embedding configuration: If you need to use a different endpoint for embeddings, configure it in the Embedding Providers (RAG使用) section. The system will automatically associate the configured embedding model when needed. If not configured, the LLM provider's base URL will be used as fallback. If neither is configured, you'll receive a clear error message with configuration instructions.

API connection errors: Check your network configuration, DNS settings, and firewall rules. Some providers may require specific network configurations.

Technical Details

This module extends Odoo's AI functionality through:

• Monkey patching of LLM API service classes
• Model inheritance for configuration management
• Provider-specific authentication and request handling
• Compatible with Odoo 19.0

Security

All API keys are stored securely in Odoo's ir.config_parameter and are only accessible to system administrators.

Support

For issues, questions, or feature requests, please contact the module author or refer to the module documentation.

Version History

• Version 1.2 - Local LLM support and enhanced configuration
  • Added Local LLM provider support (Ollama, LocalAI, etc.) - up to 3 local models
  • Independent Embedding Providers configuration table (RAG Usage)
  • Separate embedding base URL and API key configuration
  • Custom model name support for local LLMs
  • Smart model selection - only configured providers/models are shown
  • User-friendly error messages when providers are not configured
  • Multi-language support (Chinese and English) with complete UI translation
  • Improved configuration interface with better organization
  • Optimized Local LLM configuration table: Full-width layout without horizontal scrollbar, flexible column widths
  • Enhanced configuration persistence: Automatic preservation of Local LLM settings across module updates
  • Improved embedding model name handling with intelligent default value management
• Version 1.1 - Performance improvements and optimizations
  • Optimized logging levels (reduced production overhead)
  • Added function result caching for better performance
  • Improved string operations and message processing
  • Enhanced Qwen API mode detection with caching
  • Connection pool infrastructure for future optimizations
  • Code quality improvements and better error handling
• Version 1.0 - Initial release with support for DeepSeek, Qwen, ERNIE, GLM, Moonshot, and Doubao
  • Support for 6 AI providers with 20+ official API models
  • Qwen compatible mode and native API support
  • Automatic message chunking for Qwen API
  • DeepSeek embedding fallback to OpenAI
  • Performance optimizations (logging, caching, connection pooling)
  • Odoo 19 compatibility with automatic Provider structure detection
  • Comprehensive error handling and retry mechanisms

Model List

The module supports the following official API models (not open-source-only versions):

• Qwen: qwen-turbo, qwen-plus, qwen-max, qwen-max-longcontext
• GLM: glm-4, glm-4-flash, glm-3-turbo
• DeepSeek: deepseek-chat, deepseek-coder, deepseek-reasoner
• Moonshot: moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k
• Doubao: doubao-seed-1-8-251228, doubao-seed-1-6-flash-250828, doubao-seed-1-6-lite-251015, doubao-seed-1-6-251015, doubao-1-5-pro-32k-250115
• ERNIE: ernie-bot-turbo, ernie-bot, ernie-bot-4, ernie-bot-8k
• Local LLM: llama2, llama3, mistral, mixtral, qwen, codellama, phi, gemma, and custom models

Note: Only official API-available models are included. Open-source-only model sizes are not listed as they cannot be accessed via API. For Local LLM, you can use any model supported by your local LLM server (e.g., Ollama).

Multi-Language Support

This module supports the following languages with complete UI translation:

• 简体中文 (Simplified Chinese) - Full Chinese interface with all labels, help texts, and error messages translated
• English - Complete English translation for all UI elements, configuration labels, and user-facing messages

When you switch Odoo's language to English, all module-related text will automatically display in English, including:

• Configuration section titles and labels
• Field names and help texts
• Error messages and validation prompts
• Placeholder texts and tooltips