Mistral Vibe: Model Messages Before Tool Calls Now Show in the Transcript

Mistral VibeOct 30, 2026View original changelog

Mistral Vibe 3.0.4 now shows the messages the model writes before calling a tool in the transcript, where they were previously hidden. Users can read the model's stated intent for each action as the session unfolds, which makes it easier to follow and audit what the agent is doing. The change sits in the same release as a broader pass on how Vibe renders assistant output.

Key Takeaways

Pre-tool messages are no longer hidden, so the text a model writes before calling a tool now appears in the transcript.
The change gives each tool call visible context, pairing the model's stated intent with the action it takes.
Reviewing a long session is easier because the reasoning and the action sit next to each other in order.
It helps users judge tool approvals with more information about why the agent is asking.
This is a behavior change, not just styling, so transcripts may show more lines than before.
It ships in Mistral Vibe 3.0.4 alongside a broader refresh of how assistant output is rendered.

What Changed

Mistral Vibe 3.0.4 changed how the transcript treats the short messages a model writes just before it calls a tool. Until this release, that text stayed hidden and only the tool call itself appeared. Now the message is printed in the transcript, in order, alongside the tool call it introduces.

Why It Matters

Agents often explain what they are about to do in a sentence or two, such as why a file is about to be read or which command is about to run. Hiding that narration meant users saw an action with no context and had to infer the reasoning. With the narration visible, a session reads as a coherent sequence of intent followed by action. That helps when scanning a long session, when reviewing a tool approval, and when reconstructing afterwards why the agent took a particular step.

Part of a Wider Rendering Pass

The same release reworked how Vibe displays assistant output more generally, including clearer markdown, code blocks and diffs across themes, and cleaner command feedback. Showing pre-tool messages is a behavior change rather than a styling tweak: content that used to be suppressed is now part of what users see. Users who rely on terse transcripts may notice the extra lines in busy sessions.