# Mistral Remote Agents Mistral Remote Agents are cloud-based asynchronous [[AI Agents]] inside the [[Mistral Vibe]] product. They run long-horizon coding work in the background while the developer keeps doing something else — refactors, test generation, dependency upgrades — then hand the result back via the same source-control and project-tracking surfaces the developer already uses. They were introduced in April 2026 alongside [[Mistral Medium 3.5]], which powers them. ## How they work - Sessions run in Mistral's cloud, not on the developer's machine - Asynchronous by design — fire-and-forget rather than turn-by-turn - **Parallel execution** — many sessions can run at once; the developer is no longer the bottleneck on every step - **Sandboxed** — each session runs in an isolated sandbox, including broad edits and package installs - **Pull-request handoff** — agents open a PR on GitHub and notify on completion, so review happens at the result level rather than every keystroke - **Observable** — file diffs, tool calls, and progress states are inspectable mid-run - **Teleport from local CLI** — a local [[Mistral Vibe CLI]] session can be promoted to a remote one, carrying session history, task state, and pending approvals across - Powered by [[Mistral Medium 3.5]] (128B dense, 256k context) ## Intended workloads Mistral's stated sweet spot: "high-volume, well-defined work that takes a developer's time without taking their judgment." - Module refactors - Test generation - Dependency upgrades - CI investigations - Bug fixes against well-scoped issues ## Integrations - **GitHub** — code edits, branches, pull requests, and review notifications - **Linear / Jira** — issues and tickets as session inputs and outputs - **Sentry** — incident reports as triggers and sources of context - **Slack / Teams** — task notifications, status updates, approvals - "Vibe sits between the systems engineering teams already use, with humans in the loop wherever they're needed" ## Availability Pro, Team, and Enterprise plans of [[Mistral Vibe]] and [[Le Chat]] (public preview, April 2026). ## Where they fit - **Sync agentic coding** (e.g., [[Claude Code]], [[Mistral Vibe]] interactive) — turn-by-turn pair programming; you stay in the loop - **Mistral Remote Agents** — async batch / background coding; you delegate and check back later - **Multi-agent orchestration / swarms** — coordinated teams of agents on broader tasks; remote agents are typically a worker tier ## Why they matter - Pushes agentic coding from "AI in the editor" to "AI as a background worker" — the developer's role shifts from operator to dispatcher - Plays to Mistral Medium 3.5's design strengths: long-horizon tasks, multi-tool calling, structured output for downstream code consumption - Closes the loop with project-tracking systems (Linear / Jira) and chat (Slack) so async work doesn't get lost ## Caveats - Remote execution + cloud sessions raise security and compliance considerations for regulated codebases — review Mistral's data handling guarantees before pointing it at private repos - "High-volume, well-defined work" is the documented sweet spot; ambiguous or novel tasks remain better suited to interactive sessions - Public preview as of April 2026 — pricing, concurrency limits, and integration depth will change ## References - Launch post: https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5 - The New Stack coverage: https://thenewstack.io/mistral-vibe-cloud-agents/ ## Related - [[Mistral AI]] - [[Mistral Vibe]] - [[Mistral Medium 3.5]] - [[AI Agents]] - [[Agentic Engineering]] - [[Claude Code]] - [[AI Agent Orchestration]]