mistral capabilities

Canvas opens a side panel inside Le Chat for creating and iteratively editing documents, code, spreadsheets, presentations, and rich content. The AI can modify output in place across multiple drafts within the same session.

Conversational AI assistant available at chat.mistral.ai (now branded Vibe), supporting multi-turn conversations with Mistral's frontier models. Offers three modes: Chat (general conversations), Work (multi-step task delegation), and Code (developer-focused agentic coding).

Le Chat executes Python code in a secure, isolated sandbox to perform data analysis, generate charts and visualizations, do calculations, and transform files. Users describe tasks in plain language and Vibe writes and runs the code.

Vibe connects to 100+ external tools via native OAuth connectors and MCP (Model Context Protocol). Featured connectors include Atlassian (Confluence/Jira), Box, GitHub, Gmail, Google Calendar, Linear, Notion, Outlook, SharePoint, Slack, and Stripe. Knowledge connectors (Google Drive, SharePoint Online) index files for instant search.

Users can share a snapshot of any Le Chat conversation via a link or directly to social platforms. Shared conversations are cloned snapshots, not live conversations.

Users can create reusable custom agents in Le Chat by defining a name, description, and detailed system instructions. Agents can be coupled with Libraries and connectors for consistent specialized behavior (brand voice, legal review, FAQ bots, translation with glossaries).

Deep Research runs an independent, multi-step web research workflow: it builds a search plan, browses multiple sources, and synthesizes findings into a structured cited report. Users can edit the plan before execution.

Users can upload documents (PDFs, Word, spreadsheets, images, scanned documents) and photos directly into Le Chat. Mistral OCR is the default document understanding model, extracting text, tables, equations, and embedded media from complex layouts.

Flash Answers delivers text responses at approximately 1,000 words per second (over 1,100 tokens/sec), making Le Chat roughly 10x faster than comparable assistants. Powered by Cerebras Wafer Scale Engine 3 with speculative decoding.

Le Chat generates images from text prompts using Black Forest Labs FLUX Ultra (and advanced image editing capabilities developed with Black Forest Labs). Supports both photorealistic and creative/imaginative styles.

Libraries are reusable knowledge bases built from uploaded files and web pages. Attach a Library to any conversation and Vibe searches it in real time, returning answers grounded in your data with numbered source citations.

Le Chat (now Vibe) has native apps for iOS and Android that mirror the full web feature set including chat, web search, canvas, image generation, voice mode, file upload, and connectors.

Le Chat / Vibe offers five subscription tiers with escalating feature access and usage limits. Free gives broad access to core features; Pro and above expand daily quotas and add privacy controls; Enterprise enables custom models and self-deployment.

Le Chat Enterprise offers flexible deployment (cloud SaaS, dedicated VPC, private cloud, or fully self-hosted) with strict data privacy controls, SAML SSO, audit logging, and GDPR-aligned data residency options.

Projects group related chats into persistent, context-rich workspaces. Each project has its own default Library, remembers enabled tools and settings, and keeps conversations, documents, and files together.

Voice mode lets users speak instead of type. Speech is transcribed into text for review and editing before sending. Powered by Voxtral, Mistral's proprietary speech-to-text model series optimized for fast, accurate multilingual transcription.

Le Chat / Vibe can search the public web in real time and weave cited results into responses. For news queries it draws from professional partners including AFP and Associated Press for editorially vetted content.