Mistral AI Core Workflow B: Embeddings & Function Calling

Overview
Prerequisites
Instructions
Output
Error Handling
Examples
Resources

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-auth setup
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.

mistral-core-workflow-b