micsapp Mistral AI Core Workflow B: Embeddings & Function Calling
Table of Contents
Overview
Secondary workflows for Mistral AI: text embeddings for semantic search/RAG and function calling for tool-augmented interactions. Uses mistral-embed (1024 dimensions) for embeddings and mistral-large-latest for function calling.
Prerequisites
- Completed
mistral-install-authsetup - Familiarity with
mistral-core-workflow-a - Valid API credentials configured
Instructions
Step 1: Generate Text Embeddings
Use client.embeddings.create() with model mistral-embed and inputs array. Returns 1024-dimensional vectors per input text.
Step 2: Batch Embeddings
Pass multiple texts in the inputs array for efficient batch processing. Map response data array to extract embedding vectors.
Step 3: Build Semantic Search
Implement SemanticSearch class with indexDocuments() (embeds all docs) and search() (embeds query, ranks by cosine similarity, returns top-K results). Use cosine similarity: dot product divided by product of norms.
Step 4: Define Function Tools
Create tool definitions with JSON Schema parameters. Each tool has type function, name, description, and parameter schema with required fields.
Step 5: Implement Function Calling Loop
Send messages with tools and toolChoice: 'auto' to client.chat.complete(). Check for toolCalls in response. Execute matching tool function, add result as role: 'tool' message, and loop until model returns final text response.
Step 6: Build RAG Pipeline
Combine semantic search with chat completion. Retrieve relevant documents for user query, inject as context in system prompt, generate response with mistral-small-latest. Instruct model to answer from context only.
Output
- Text embeddings with
mistral-embed(1024 dimensions) - Semantic search with cosine similarity ranking
- Function calling with tool execution loop
- RAG chat combining retrieval and generation
Error Handling
| Issue | Cause | Resolution |
|---|---|---|
| Empty embeddings | Invalid input text | Validate non-empty strings before API call |
| Tool not found | Unknown function name | Check tool registry matches definitions |
| RAG hallucination | Insufficient context | Add more documents, tune retrieval top-K |
| High latency | Large batch size | Split into smaller batches, add concurrency |
Examples
Embeddings
const response = await client.embeddings.create({
model: 'mistral-embed',
inputs: ['Machine learning is fascinating.'],
});
console.log(`Dimensions: ${response.data[0].embedding.length}`); // 1024 # 1024: 1 KB
Function Calling
const response = await client.chat.complete({
model: 'mistral-large-latest',
messages: [{ role: 'user', content: 'Weather in Paris?' }],
tools, toolChoice: 'auto',
});
See detailed implementation for advanced patterns.