Mistral Vibe: Image Attachments for Vision-Capable Models
Mistral VibeMistral Vibe v2.14.0 introduced image attachments via @-mentions in the TUI, enabling users to reference images directly in the chat interface when working with vision-capable models. Users can now @-mention an image file in the prompt input and the model will receive and reason over it inline. This removes the need for workarounds such as describing images in text or pre-processing them separately, making multimodal workflows a first-class experience in the terminal.
Image Attachments in the TUI
Mistral Vibe v2.14.0 brings image attachments to the terminal UI, allowing users to reference image files directly through @-mention syntax when working with vision-capable models.
How It Works
In any session backed by a vision-capable model, users can now type @ in the chat input to bring up the file picker and select an image file. The selected image is attached to the prompt and sent to the model as part of the conversation turn. The model receives the image data inline and can reason over its contents, answer questions about it, describe it, or use it as context for code generation or other tasks.
Why This Matters
Until this release, working with images in Mistral Vibe required manual workarounds: converting images to base64, describing them in text, or using external tools to pre-process visual content before passing it to the agent. Image attachments via @-mentions eliminate this friction entirely. The workflow now matches what users expect from a modern multimodal assistant, and it fits naturally into the existing @-mention file-reference pattern that Mistral Vibe users already know.
Compatibility
The feature is gated on model capability: it activates when the active model supports vision input. Users working with text-only models will not see image files offered in the @-mention picker for this purpose.
Bug Fixes and Reliability
Alongside image attachments, v2.14.0 also improved LLM call reliability by adding automatic retries on network errors and timeouts, including connection failures, read/write errors, remote protocol errors, and timeout events. This means temporary network blips no longer abort an in-progress agent run.