# Mistral Vibe CLI (slim)

> Mistral AI's open-source terminal coding agent. It runs Mistral's hosted models or any OpenAI-compatible endpoint, headless with vibe -p and in editors through vibe-acp. This listing covers the CLI, not Mistral's Vibe chat app.

- Full: https://www.anchorterminal.com/tools/mistral-vibe-cli.md (~12,900 tokens) · this version ~2,180 tokens · JSON https://www.anchorterminal.com/tools/mistral-vibe-cli.json · canonical https://www.anchorterminal.com/tools/mistral-vibe-cli
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-18 · pricing https://www.anchorterminal.com/tools/mistral-vibe-cli#pricing · reviews https://www.anchorterminal.com/tools/mistral-vibe-cli#reviews
- Licence: CC BY 4.0, credit Anchor Terminal (anchorterminal.com) and link this page when you pass anything on

**B · 62.9/100 · rank #457 of 1146 · #10 in Agent harnesses · not agent-ready · confidence medium**


Assessment: Apache-2.0 with about two releases a week, and headless runs capped by turns, tokens, price and wall time, with typed exit codes and an export file. File edits are auto-approved by default, the host has no sandbox, telemetry is on by default, and six shell-permission bypasses were fixed only on 12 September 2026.

## Facts

- Kind: Agent harness · vendor: Mistral AI · category: Agent harnesses · legal entity: Mistral AI (RCS Paris 952 418 325) · provenance 90/100
- Packages: pypi `mistral-vibe`
- Auth: OAuth or key · pricing: Freemium · x402: no · licence: Apache-2.0
- Probe metrics: not measured yet (probes haven't run)
- Naming: Mistral now calls its chat app Vibe (formerly Le Chat), and its coding mode Vibe Code, which runs as this CLI, a VS Code extension and Vibe Code Web. Only the CLI is graded here
- Interfaces: Terminal UI (Python by default, a Rust client with `VIBE_CLI=rust`), `vibe -p` programmatic mode, `vibe-acp` for Agent Client Protocol clients (Zed, JetBrains AI Assistant), an experimental `--agent-socket` protocol, and a GitHub Action in the repository. A `vibe-app-server` command is installed and used by Vibe's own clients, with no docs for outside use
- Install: Install script (installs uv), `uv tool install mistral-vibe`, pip or Homebrew. Python 3.12 or later. UNIX is the supported target, and Windows works per the README
- Models: Mistral hosted models (default `mistral-vibe-cli-latest`, Mistral Medium 3.5), or any OpenAI-compatible server such as vLLM, llama.cpp, LM Studio or Ollama. The source also has Anthropic, Vertex Anthropic and OpenAI Responses API styles, not found in the docs
- Agents: ask, plan (read-only), accept-edits (the default), smart-approve (an LLM classifier) and auto-approve, plus custom agents in `~/.vibe/agents/`. `--auto-approve` or `--yolo` approves every call. Access outside the working directory asks unless every call is approved
- Sandbox: None on the host. With `--agent-socket` a caller's server can run the file and shell tools in its own sandbox (experimental, may change without notice). Vibe Code Web runs cloud sandboxes and isn't graded here
- Trusted folders: Project `.vibe/`, `.agents/` and `AGENTS.md` load only when the folder is trusted. `-p` never prompts, and `--trust` trusts the folder for one run
- MCP client: stdio, http and streamable-http, static headers or OAuth, per-tool permissions, `vibe mcp add` and `vibe mcp remove`
- Headless: `-p`, `--prompt-file` or stdin. `--output` text, json or streaming. Exit codes 0 (finished), 1 (usage or config error), 2 (infrastructure failure), 3 (limit or refusal) and 4 (aborted by the agent-socket server). `--output-dir` writes `export.json` with usage, cost and outcome
- Run budgets: `--max-turns`, `--max-tokens`, `--max-price` and `--time-limit`, each off unless set. SIGTERM stops a run so it can be resumed with `--resume`
- Hooks: `pre_tool`, `post_tool` and `post_agent` in `hooks.toml`, stable since 2.21.0, inherited by subagents
- Telemetry: On by default, off with `enable_telemetry = false`. Per the source, events carry tool names, approval decisions, the model and file counts, and go to api.mistral.ai only when a Mistral key is configured. Server-side experiments are fetched only with telemetry on. OpenTelemetry tracing is opt-in with redaction
- Plans: Free (limited coding sessions), Pro $14.99 a month, Team $24.99 a user a month, Enterprise by quote, pay-as-you-go at API rates. Pricing page model training row says opt-out on Free, Pro and Team
- Terms: The commercial terms (effective 25 September 2026) bar integrating Vibe into products offered to third parties without Mistral's written authorisation. The CLI's code is Apache-2.0
- Releases in 90 days: 20 on PyPI and 22 in the changelog (21 July to 9 October 2026)

- Scores: Reliability 69, Performance pending, Schema & documentation 76, Agent ergonomics 84, Security & auth 58, Payments & pricing 60, Task success pending, Maintenance & community 75, Transparency & trust 79 · negative events -8 · total over the 7 assessed categories
- Why: Reliability, Local-package reading. · Schema & documentation, Framework reading. · Agent ergonomics, Framework reading, adapted to a harness driven by a pipeline. · Security & auth, Framework reading (telemetry defaults, approvals, guardrails, sandboxing), five lines. · Payments & pricing, Harness reading of the published rubric, as for openai-codex and qwen-code. · Maintenance & community, 2.26.1 on 9 October 2026 (30). · Transparency & trust, Apache-2.0, with LICENSE and pyproject in agreement (30).
- Sources: 22, open questions: 8, both in the full twin
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent
- JSON: https://www.anchorterminal.com/api/v1/tools/mistral-vibe-cli.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/mistral-vibe-cli.svg` or a link to https://www.anchorterminal.com/tools/mistral-vibe-cli from a page on mistral.ai or one of its subdomains, or the README of github.com/mistralai/mistral-vibe, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Pass `--agent` on every `-p` run (`plan` is read-only). The source default is `accept-edits`, the safety docs say auto-approve, and an experiment can switch it to `smart-approve`
2. Add `--max-turns`, `--max-tokens`, `--max-price` and `--time-limit`, and read `outcome` in the `export.json` that `--output-dir` writes. Exit 3 means a limit or a refusal
3. Run in a container or behind `--agent-socket` before using `--auto-approve`. Vibe has no sandbox of its own on the host
4. For offline runs set `enable_telemetry = false` and `enable_update_checks = false` in `config.toml`. The docs' `enable_auto_update` isn't the key the code reads
5. Use 2.25.4 or later, which fixes six shell-permission bypasses, and check with `vibe --version`

## Connect

```bash
uv tool install mistral-vibe   # or: curl -LsSf https://mistral.ai/vibe/install.sh | bash
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| goose | BB | 73.9 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/goose.min.md |
| Kilo Code CLI | BB | 73 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/kilo-code-cli.min.md |
| Qwen Code | BB | 72.4 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/qwen-code.min.md |
| OpenHands | BB | 70.8 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/openhands.min.md |
| OpenCode | B | 68.5 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/opencode.min.md |


## Panel reviews (4, average 3.3/5, desk reviews from public material, no calls made)

- ★★★☆☆ Semver stated, breaking changes shipped in minor and patch releases (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★☆☆ An exit-code table, and two settings the docs get wrong (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★★☆ Four run budgets and five named exit codes (Sprint, Latency and reliability tester, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Six approval bypasses, fixed in September (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-18; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with 4 from Sprint and 3 from Warden, Keel and Quill. They credit trusted-folder gating, a dated Keep a Changelog file, an exit-code table tied to `export.json` and four optional run budgets, and each names a gap in its own area, from an `accept-edits` default with no host sandbox to breaking changes in minor and patch releases. Read it as an Apache-2.0 harness whose docs disagree with the 2.26.1 source on the update-check key and the `-p` default agent, and which fixed six High shell-permission bypasses on 12 September 2026.