Knowledge Retriever Tool

• Semantic Search: Find relevant documents using natural language queries
• Source Citations: AI responses include references to source documents
• Hybrid Search: Combine semantic and keyword search for better accuracy
• Collection-Aware: Search specific collections or across all knowledge
• Configurable Relevance: Set minimum similarity thresholds

Dual Integration

User: "What's our refund policy?"
AI: Uses knowledge_retriever tool → searches policy documents → cites sources

Claude Desktop → MCP Server (Odoo) → knowledge_retriever → returns relevant docs

For Odoo AI Assistants

1. Go to LLM → Assistants → Assistants
2. Open or create an assistant
3. Navigate to the Tools tab
4. Enable the "knowledge_retriever" tool
5. The assistant can now search knowledge collections

For External MCP Clients

1. Install and configure llm_mcp_server module
2. The knowledge_retriever tool is automatically exposed via MCP
3. Configure your MCP client (Claude Desktop, etc.) to connect to Odoo
4. External AI can now search your knowledge base

Example 1: Odoo Assistant with Knowledge Access

Setup:

• Create knowledge collection with company policies
• Enable knowledge_retriever tool on support assistant

Result:

User: "What's the warranty period for laptops?"
Assistant: [Searches policies collection]
"Based on our Electronics Warranty Policy, laptops have a 2-year warranty covering
hardware defects. (Source: Electronics Warranty Policy, updated Jan 2024)"

Example 2: Claude Desktop Accessing Odoo Knowledge

Setup:

• Configure llm_mcp_server with your Odoo instance
• Add server to Claude Desktop MCP settings
• Index product documentation in Odoo knowledge

Result:

Claude Desktop → uses knowledge_retriever tool → searches Odoo docs
Returns: Relevant product specs with source citations from your knowledge base

Example 3: Continue.dev with Company Codebase

Setup:

• Index code documentation in Odoo knowledge collection
• Expose via MCP server
• Configure Continue.dev to use Odoo MCP server

Result:

Developer asks Continue.dev about internal APIs → searches indexed docs → provides accurate answers from your actual documentation.

Tool Input Schema

{
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "description": "Search query to find relevant knowledge"
    },
    "collection_id": {
      "type": "string",
      "description": "ID of knowledge collection to search (optional)"
    },
    "top_k": {
      "type": "integer",
      "description": "Number of results to return (default: 5)"
    },
    "min_similarity": {
      "type": "number",
      "description": "Minimum similarity score 0-1 (default: 0.7)"
    }
  },
  "required": ["query"]
}

Tool Execution Flow

1. Receive query: AI assistant or external tool calls knowledge_retriever
2. Vector search: Query is embedded and searched against knowledge chunks
3. Filter results: Apply similarity threshold and top_k limit
4. Return sources: Chunks with metadata, similarity scores, and source references
5. AI uses context: Assistant incorporates results into response with citations

Tool Registration

Defined in data/llm_tool_data.xml:

<record id="llm_tool_knowledge_retriever" model="llm.tool">
    <field name="name">knowledge_retriever</field>
    <field name="description">Retrieves relevant knowledge from document database
    using semantic search...</field>
    <field name="implementation">knowledge_retriever</field>
    <field name="active" eval="True" />
</record>

Implementation

Located in models/llm_tool_knowledge_retriever.py:

class LLMToolKnowledgeRetriever(models.Model):
    _inherit = "llm.tool"

    @api.model
    def _get_available_implementations(self):
        implementations = super()._get_available_implementations()
        return implementations + [("knowledge_retriever", "Knowledge Retriever")]

Search Process

1. Embed query using collection's embedding model
2. Perform vector similarity search in vector store (pgvector/Qdrant/Chroma)
3. Filter by min_similarity threshold
4. Return top_k most relevant chunks
5. Include source document metadata

Internal Odoo Assistants

• Customer Support: Search FAQ, policies, product docs
• HR Assistant: Search employee handbook, HR policies
• Sales Assistant: Search product specs, pricing, competitor analysis
• IT Helpdesk: Search technical documentation, troubleshooting guides

External MCP Integration

• Developer Tools: Continue.dev, Cursor accessing code documentation
• Claude Desktop: Personal assistant with access to company knowledge
• Custom AI Apps: Build external apps that query Odoo knowledge
• Multi-Tool Workflows: Chain knowledge search with other MCP tools

Access Control

• Tool: Requires llm.group_llm_user to execute
• Collections: Respects Odoo record rules and access rights
• MCP Server: Separate authentication for external access

Data Privacy

• Knowledge searches respect user permissions
• External MCP access requires explicit configuration
• No knowledge is shared unless explicitly indexed in collections

Tool not appearing

1. Verify module is installed and active
2. Check llm_knowledge module is installed
3. Refresh assistants or MCP client

Search returns no results

1. Check collection has processed resources (state=ready)
2. Verify embeddings are generated
3. Lower min_similarity threshold
4. Check vector store is configured correctly

MCP connection fails

1. Verify llm_mcp_server is installed and configured
2. Check MCP client configuration matches Odoo URL
3. Review authentication credentials
4. Check Odoo is accessible from MCP client network