Mistral Vibe: /retry Command and Visible LLM Retries

Mistral VibeView original changelog

Mistral Vibe v2.23.3 introduced a new /retry command that lets users retry an interrupted response without retyping their last message or restarting the turn. The same release also made automatic LLM retries visible in the UI and logs, so transient provider failures that the agent silently recovers from are no longer hidden from the user. Together, the two changes give developers much clearer insight into β€” and control over β€” what happens when a model call fails or a turn gets cut short.

Key Takeaways

  • Mistral Vibe added a /retry command that resumes an interrupted response without forcing the user to retype their prompt.
  • Automatic LLM retries triggered by transient provider failures are now shown live in the terminal UI instead of happening silently.
  • Retry activity is also written to the session logs, making it easier to diagnose flaky connections or rate-limiting after the fact.
  • The change addresses a long-standing pain point where sessions appeared to "hang" with no explanation during a provider hiccup.
  • Both features shipped together in v2.23.3, released August 3, 2026.
  • The improvements are aimed at developers running longer or higher-volume agent sessions where transient failures are more likely to occur.

A Dedicated Command for Interrupted Turns

Mistral Vibe v2.23.3 introduced a new /retry command that lets a user retry an interrupted response directly, without having to retype the last message or restart the entire turn from scratch. Interruptions can happen for a range of reasons β€” a dropped connection, a manually cancelled request, or a transient provider error β€” and previously the only recovery path was to re-enter the prompt and hope the agent picked up where it left off. With /retry, Mistral Vibe now resumes the same turn cleanly, saving time and avoiding the risk of duplicated or inconsistent instructions creeping into the conversation history.

LLM Retries Are No Longer Invisible

The release also surfaced Mistral Vibe's internal LLM retry behavior directly in the UI and logs. Mistral Vibe has long retried failed model calls automatically behind the scenes, but until now those retries were effectively invisible β€” a session could appear to simply "hang" for a few seconds with no indication of why. Now, when the underlying provider call fails and Mistral Vibe retries it, that activity is reflected in real time in the terminal UI and recorded in the session logs. This gives developers a much better mental model of what the agent is actually doing during moments of apparent inactivity, and makes it far easier to diagnose flaky provider connections or rate-limit issues when they occur.

Why It Matters

Both changes target the same underlying pain point: turns that fail or stall without clear recourse or explanation. Combined, they make Mistral Vibe noticeably more resilient and transparent when working against real-world, occasionally unreliable network and provider conditions β€” a meaningful quality-of-life improvement for anyone running longer, unattended, or high-volume agent sessions.