Conversational interfaces are great for creating drafts, but professionals still spend hours refining their work elsewhere—copying drafts into editors, tweaking text, and juggling scattered prompts.
At Mistral AI, we built Canvas to eliminate this friction. It merges drafting, refining, and finalizing into one seamless space where AI assistance and manual editing coexist.
How it works
•
Generate & Edit
Manually edit generated content without losing context.
•
Highlight & Improve
Reference any section to condense, expand, or refine—while maintaining control over the final result.
•
Prototype in Code
Generate and edit React code in a version-controlled editor and instantly preview changes.
By integrating LLMs with manual editing, Canvas reduced both prompts and completion time by 50%, shifting user focus from process to quality.
AI Interaction
•
Direct Content Manipulation
One of the first approaches to integrate direct manipulation of generated content into chat UIs.
•
Right-click menus and inline actions
Right-click menus and inline actions based on familiar see and point interactions from GUIs.
•
Referencing on-screen content
Referencing on-screen content reduces the need for detailed prompts in a "what you see is what you get" manner.
Canvas is one step toward more intuitive AI interfaces. Future human-AI collaboration will go further, anticipating intent, adapting to context in real time, and integrating AI into familiar interaction patterns.