Built by Metorial, the integration platform for agentic AI.
Generate a chat completion using Mistral AI models. Supports multi-turn conversations with system, user, assistant, and tool messages. Can use function calling, structured JSON output, and safety prompts.
Generate dense vector embeddings for text or code. Useful for semantic search, retrieval-augmented generation (RAG), clustering, and content similarity comparison. Supports both text embeddings (mistral-embed) and code embeddings (codestral-embed).
Generate code using fill-in-the-middle (FIM) completion. Provide a code prompt and optional suffix to generate code that fits between them. Ideal for code insertion, auto-completion, and infilling tasks.
Analyze plain text or chat messages for harmful content across multiple safety categories. Returns boolean flags and confidence scores for categories including sexual content, hate/discrimination, violence, dangerous content, self-harm, health/financial/legal misinformation, and PII detection.
Send messages to a Mistral AI agent and get a completion. Agents combine language models with built-in connectors for code execution, web search, image generation, and document libraries. Agents are created in the Mistral AI console.
List files uploaded to the Mistral AI platform. Files are used for fine-tuning datasets, batch inference inputs, and OCR processing.
Create a fine-tuning job to customize a Mistral model on your training data. Upload JSONL training files first using the file management tools, then reference their IDs here. Supports configurable hyperparameters including learning rate, training steps, and epochs.
Delete a file from the Mistral AI platform. This permanently removes the file and cannot be undone.
Cancel a running or queued fine-tuning job. Cannot cancel already completed or failed jobs.
List all fine-tuning jobs in your workspace. Returns job statuses, models, and configuration details.
Cancel a running or queued batch inference job.
List batch inference jobs. Optionally filter by status.
Extract text, tables, and images from documents using Mistral OCR. Supports PDFs and images. Returns structured content in markdown format with optional table formatting (markdown or HTML). Can extract headers, footers, and hyperlinks.
List available Mistral AI models including both Mistral-provided and user fine-tuned models. Returns model details such as capabilities, context length, aliases, and ownership.
Retrieve the current status and details of a fine-tuning job. Use this to monitor training progress, check for completion, or get the resulting fine-tuned model ID.
Create a batch inference job for processing multiple requests asynchronously at reduced cost. Upload a JSONL input file first, then reference it here. Supports chat completions, embeddings, FIM, moderation, OCR, and more.
Retrieve the status and details of a batch inference job. Use this to monitor progress, check completion, and get output/error file IDs.
Delete a fine-tuned model from your Mistral AI workspace. This permanently removes the model and cannot be undone. Only works on user-created fine-tuned models, not Mistral-provided models.
Retrieve metadata and a signed download URL for a file uploaded to Mistral AI.
Retrieve metadata for a specific Mistral AI model, including capabilities, context length, aliases, ownership, and fine-tuned model details when available.
Transcribe an audio file with Mistral AI's audio transcription API. Supports public file URLs, Mistral file IDs, or inline base64 audio, with optional language hints, diarization, context bias, and timestamps.
Download file content from Mistral AI and return it as a Slate attachment. Use this for batch output/error files, uploaded documents, or other downloadable files.
Generate speech audio from text using Mistral AI. Returns the generated audio as a Slate attachment instead of inline base64.
List available Mistral AI speech voices, excluding sample audio data. Use a returned voiceId with Generate Speech when a voice is required.
Upload a file to Mistral AI for OCR, batch inference, fine-tuning datasets, or later transcription by file ID. Accepts plain text content or base64-encoded bytes.