Mistral AI
AIRun the full Mistral AI platform in your flows: chat and vision, Document AI OCR, embeddings, code completion, moderation, audio, agents, and batch processing.
What This Integration Enables
Mistral gives an agent a broad, coherent toolkit from a single provider. Chat and vision handle reasoning and images, Document AI does OCR on scanned files, embeddings power retrieval, code models handle fill-in-the-middle, moderation screens content, and batch processing runs large jobs cheaply. Agents, conversations, and libraries support stateful, multi-turn work. Breadth from one vendor means fewer integrations to maintain and a consistent interface across steps. What makes it operational is the layer that sequences these capabilities and inserts a person at the review point. An orchestration layer chains OCR into summarization into filing, and pulls legal in on the clause that needs a human. FlowRunner is built for that layer.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Contract Review Pipeline
A contract PDF arrives, often as a scan. The agent runs Document AI OCR to read it into structured text, asks a chat model to summarize the terms and flag anything non-standard, and files the structured summary against the record. Legal reviews a summary with the flagged clauses highlighted instead of reading every page, and any unusual clause routes to a lawyer before the contract advances.
Code-Assisted Operations
An internal tool needs small code changes and script generation on a schedule. The agent uses Mistral's code models to draft the changes with fill-in-the-middle, and a person reviews the diff before it merges. Routine code work speeds up while a human still owns what ships.
Moderated Content Intake
User-generated content needs screening and classification at volume. The agent runs Mistral moderation on each submission and classifies the clean ones automatically. Anything the moderation model flags is routed to a human moderator with the model's read attached, so the borderline calls are made by a person.
Human-in-Loop Highlight
A model that summarizes a contract or drafts code produces work a person still needs to own. When Mistral flags a non-standard clause, drafts a change that will ship, or moderation marks content as borderline, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the output and its context, and sends it to the right person via Slack. They review and decide. The full Mistral surface does the work; a person owns the consequential calls.
Agent Capabilities
37 actionsChat
3- Ask AI Sends a single prompt to a Mistral chat model and returns the generated answer text. Supports an optional system prompt, sampling controls (temperature, top-p), token limits, deterministic seeding, guardrail injection, JSON output mode, and stop sequences.
- Create Chat Completion Runs a full multi-message chat completion against any Mistral chat model and returns the raw API response, including choices and token usage.
- Analyze Image Analyzes one or more images with a multimodal Mistral model (e.g. mistral-medium-latest, mistral-small-latest, pixtral) and answers a prompt about them.
Document AI
1- OCR Document Extracts text and structure from a PDF or image using Mistral Document AI (OCR). Returns per-page markdown, optional embedded images as base64, and optional paragraph-level content blocks with bounding boxes.
Embeddings
1- Create Embeddings Generates vector embeddings for one or more input texts using a Mistral embedding model ('mistral-embed' for natural language, 'codestral-embed' for code).
Code
1- FIM Completion Performs a fill-in-the-middle code completion with a Codestral model: given a code prompt (prefix) and an optional suffix, generates the code that belongs between them.
Moderation
2- Moderate Text Classifies one or more raw texts for harmful content with Mistral's moderation model. Each result contains boolean flags and confidence scores across nine safety categories: sexual, hate_and_discrimination, violence_and_threats, dangerous_and_criminal_content, selfharm, health, financial, law, and pii.
- Moderate Conversation Classifies a chat conversation for harmful content with Mistral's conversational moderation endpoint, which assesses the last message in the context of the full conversation.
Audio
2- Transcribe Audio Transcribes an audio file into text using Mistral's Voxtral speech-to-text models. Accepts a public audio URL (including FlowRunner file URLs) or a previously uploaded Mistral file ID.
- Text to Speech Converts text into natural-sounding speech using Mistral's Voxtral TTS model, with support for preset and custom (cloned) voices.
Files
6- Upload File Uploads a file from FlowRunner file storage (or any accessible URL) to Mistral for later use in OCR, batch processing, fine-tuning, or audio transcription.
- List Files Lists files uploaded to your Mistral AI account with pagination, optional name search, and purpose filtering.
- Get File Retrieves metadata about a single file uploaded to Mistral, including its name, size, purpose, source, and creation timestamp.
- Delete File Permanently deletes a file from your Mistral AI account. Returns a confirmation object indicating whether the deletion succeeded.
- Get File Signed URL Generates a temporary signed download URL for a file stored on Mistral (e.g. batch job results).
- Download File Downloads a file's content from Mistral (e.g. batch job output) and saves it to FlowRunner file storage.
Batch
4- Create Batch Job Creates an asynchronous batch job that processes a . jsonl file of API requests (uploaded with the 'batch' purpose) at a 50% discount compared to synchronous calls.
- List Batch Jobs Lists batch jobs in your Mistral AI organization with pagination and optional status filtering.
- Get Batch Job Retrieves the current state of a batch job, including its status, request counters (total, completed, succeeded, failed), timestamps, and the output/error file IDs once finished.
- Cancel Batch Job Requests cancellation of a queued or running batch job. Requests already completed before cancellation remain available in the job's output file.
Agents
5- Create Agent Creates a reusable Mistral agent that bundles a model, system instructions, built-in tools (web search, code interpreter, image generation, document libraries), and completion settings into a single versioned configuration addressable by agent ID in conversations.
- Update Agent Updates an existing Mistral agent's name, model, instructions, or description. Each update creates a new agent version; only the provided fields are changed.
- List Agents Lists agents created in your Mistral AI workspace with pagination. Returns each agent's ID, name, model, tools, and version information.
- Get Agent Retrieves the full configuration of a single Mistral agent, including its model, instructions, tools, completion settings, and version history metadata.
- Delete Agent Permanently deletes an agent from your Mistral AI workspace. Existing conversations that referenced the agent are not deleted.
Conversations
6- Start Conversation Starts a new server-side conversation with either a base model or an existing agent. The conversation history is stored by Mistral (unless storing is disabled) so follow-up messages can be appended later by conversation ID.
- Append to Conversation Appends a new user message to an existing stored conversation and runs the completion (including any agent tool executions).
- Get Conversation Retrieves metadata about a stored conversation, including its name, description, associated model or agent, and timestamps.
- List Conversations Lists stored conversations in your Mistral AI workspace with pagination, newest first. Returns each conversation's ID, name, model or agent, and timestamps.
- Get Conversation History Retrieves all entries of a stored conversation in order, including user messages, assistant messages, tool executions, and agent handoffs.
- Get Conversation Messages Retrieves only the user and assistant messages of a stored conversation in order, omitting tool executions and other non-message entries.
Libraries
4- Create Library Creates a document library that agents can search through the built-in document library tool (retrieval-augmented generation).
- List Libraries Lists all document libraries you created or that were shared with you, including document counts, total sizes, and ownership information.
- Upload Library Document Uploads a document from FlowRunner file storage into a Mistral document library, where it is automatically processed and indexed for agent retrieval.
- List Library Documents Lists documents stored in a Mistral document library with pagination and optional name search.
Models
2- List Models Lists all Mistral AI models available to your account, including chat, code, embedding, OCR, moderation, and audio models, with their capabilities and context window sizes.
- Get Model Retrieves detailed information about a single Mistral AI model, including its capabilities (chat, vision, function calling, fine-tuning), context window size, and aliases.
Start building with Mistral AI
$100 in credits. No card required. Connect in minutes.