Mistral Vibe 3.0.1 lowers Python requirements after the runtime overhaul
Mistral has published Vibe 3.0.1, a stable update following the previous day’s removal of the coding assistant’s legacy runtime.
A major migration followed by a broad patch
GitHub records version 3.0.0 as published on October 19 at 16.01 UTC and version 3.0.1 on October 20 at 13.55 UTC. The newer release’s pull request opened at 13.29 UTC and merged at 13.53 UTC. Its comparison with 3.0.0 contains one public release commit. That establishes the publication sequence, without dating all the work that preceded it.
Version 3.0.0 removed the old execution runtime and Python terminal interface. The command line now uses the Rust interface. The Unified Harness became mandatory across the command line, desktop, app server and editor integrations. Plan mode was removed. Vibe 1.x sessions are no longer supported, while read only imports remain for session folders from Vibe 2.0 onward. Stricter certificate validation could also reject some private certificate authorities.
The transition had already begun. Version 2.26.1 documented the Rust terminal alongside a way to return to the Python interface. Version 2.26.0 had made the Unified Harness the default and stopped silently falling back if it was unavailable. The major release closed those remaining escape routes.
Python support and administrator controls
The tagged package configuration now requires Python 3.12 or newer, down from Python 3.14 in version 3.0.0. The updated installation guidance also specifies a build with the Global Interpreter Lock enabled. Python remains part of the package requirements even though its terminal interface is written in Rust.
A new enable_auto_approve setting gives administrators a way to forbid the mode that automatically approves tool calls. When managed configuration disables it, agents using that mode are unavailable and the corresponding launch flags are ignored. Permissions for individual tools still apply. The configuration schema leaves the setting enabled by default, so this is an available control rather than a universal change in approval behavior.
The patch also fixes prompts using corporate proxies trusted only by operating system certificates when enable_system_trust_store is enabled.
More control over unfinished agent work
Experimental app server methods can wait for, interrupt and inspect subagents. Programmatic runs now wait for remaining subagents before exiting, and stops interrupt them. Exports include subagent status, usage and journals.
The tagged README also describes an experimental run server that shares app server processes across multiple runs. Each run requires a sandbox provided through an agent socket. The command is hidden from help, is unsupported on Windows and may change or disappear. Those limits matter for anyone considering it as infrastructure for repeated coding tasks.
Desktop sessions gain scheduled prompts. Recurring prompts already appeared in the previous terminal documentation.
Source publication and installation remain separate checks
At the October 20 check at 15.10 UTC, GitHub listed nine uploaded ACP release archives for version 3.0.0, covering macOS, Linux and Windows. They were added after the initial publication. Version 3.0.1 now also lists nine ACP archives across the same platforms. Automatically generated source archives are separate from those uploaded assets.
Package availability was not independently verified, and no installation or runtime tests were performed. The published source and release notes document the changes, while successful installation and behavior in a particular environment still require verification.
Illustrative archival photograph by Christina Morillo under CC0.
