raven

raven [-v]raven web [--port N] [--stop] [--foreground]raven tui [-w dir] [--home dir] [--standalone]raven onboard [options]raven agent -m message [options]raven serve [--port N] [--open]raven doctor [--probe] [--json] [--fix]raven statusraven upgrade [--check]

raven is EverMind's host-agent CLI. It orchestrates built-in agents (research, code, design, on-call) and third-party agents from one install. The console script is `raven` (`raven.cli.commands:run`); `python -m raven` is the same entry. Settings, memory, and runtime state live under ~/.raven unless RAVEN_HOME points elsewhere. Config is $RAVEN_HOME/config.json.The usual first-run path is the official installer (which places raven on PATH via uv tool), then raven onboard, then raven web or raven tui. Further command groups cover providers, plugins, skills, sessions, channels, cron, MCP, sandbox, and related operations (`raven --help` lists them).Raven is pre-alpha: interfaces and configuration may change. It needs Python 3.12 or newer. The TUI bundles a Node.js UI and expects Node.js 22+.

-v, --version

Print the Raven version and exit

Open the local page and keep a gateway running after the terminal closes. Default port 18792. --stop shuts the resident gateway down. --foreground holds the terminal and does not restart on failure. With no subcommand, bare raven does the same when a browser can be opened

Launch the native Ink/React terminal UI. --workspace/-w sets the working directory (default: current directory). --home sets the agent home. --standalone runs an embedded engine even if a gateway already hosts the page. --color is auto, truecolor, 256, 16, or none. On a headless machine, bare raven starts the TUI instead of the WebUI

Interactive wizard for the first provider, sandbox, channels, memory, and related extras. --provider, --api-key, --model, and --base-url skip prompts. --non-interactive requires flags for any missing field. --reset re-runs the wizard over an existing config. -y/--yes skips confirmations

One-shot turn; interactive chat lives in raven tui. --message-file reads the prompt from a file. -c/--continue continues the most recent CLI session. -r/--resume resumes by id or unique prefix. -w/--workspace and --home match tui. --permission-mode is ask, smart, or full

Run the headless gateway in the foreground (WebSocket RPC, plus the page when one is built). --port prefers 18792 and probes forward if taken. --open opens a browser

Report install and config health. --probe sends a test LLM message. --json prints machine-readable output. --fix applies config repairs the report knows how to make

Print the config path, workspace, default model, and which providers are configured

Install the newest release this install is entitled to (stable, or beta if this install joined that channel). --check reports without installing. Does not update an editable source checkout — use git pull and ./install.sh there. Stop raven web first, then start it again after the upgrade

~/.raven/

Default home (override with RAVEN_HOME): config, workspace, logs, memory, and runtime state

Provider credentials, default model, language, and related settings written by raven onboard and the WebUI

This raven is EverMind's agent harness, not Sentry's deprecated Python client of the same name. Distro packages named raven are usually that older library. The official installer is a curl-to-shell script that installs via uv; raven upgrade only works on a uv-managed install. LLM calls need a configured provider key. raven agent without -m or --message-file is not the interactive chat — use raven tui. A gateway started with raven serve or raven web --foreground is not restarted by the in-page updater.

Developed by EverMind AI. Apache License 2.0. The published Python package reports version 0.2.3 (2026).

claude(1), codex(1), aider(1), opencode(1), llm(1)

Source code · Homepage · Documentation