mistral-vibe

Mistral Vibe is an open-source command-line coding assistant powered by Mistral's models. It offers a conversational interface to a codebase, letting users explore, modify and interact with projects in natural language through a set of built-in tools. Installation is available via a one-line script for Linux/macOS, or through uv and pip (`uv tool install mistral-vibe`, `pip install mistral-vibe`). It requires Python 3.12+ and a modern terminal emulator; Windows works but UNIX environments are the officially supported target. Core capabilities include reading, writing and patching files, executing shell commands with managed sessions, recursive code search with grep/ripgrep, a todo list for tracking agent work, interactive clarifying questions, and task delegation to subagents. The agent automatically scans project structure and Git status to build context. Built-in agent profiles cover different approval levels: `ask` (approval for tool executions), `plan` (read-only exploration), `accept-edits` (default, auto-approves file edits) and `auto-approve` (approves everything). A default agent can be set in config.toml, and custom agents can be defined in `~/.vibe/agents/` or a project's `.vibe/agents/`. Subagents such as `explore` run independently for specialized work, with a status list showing their state and context usage. The CLI supports slash-command and file-path autocompletion, `@` mentions for files and image attachments sent to vision-capable models, persistent history, themes, multi-line input, external editor editing, and toggles for tool output, todo view and a debug console. A trust folder system asks for confirmation before running in new directories and remembers trusted folders. Programmatic mode (`--prompt`) enables scripting with options such as `--max-turns`, `--max-price`, `--max-tokens`, `--agent`, `--enabled-tools`/`--disabled-tools` (exact, glob or regex patterns) and output formats text, json or streaming. An experimental voice mode allows dictation via microphone. A skills system following the Agent Skills specification lets users add reusable components, custom slash commands and tools via `SKILL.md` files with YAML metadata, discovered from project and global directories and controllable through enable/disable patterns. Configuration lives in `config.toml` (project or user level) and covers API keys, custom domains, TLS/system trust store, OpenTelemetry tracing with redaction modes, custom system prompts and AGENTS.md instructions, MCP server configuration, session management and update settings.