hermes-beaver

A lightweight, full-screen dashboard that runs next to Hermes Agent and shows what it is doing:

• current tasks and the queue (GitHub, Telegram, direct);
• what needs your feedback, and recently finished work;
• CPU, memory, disk and GPU load, with 30-minute sparklines;
• an ASCII beaver whose mood follows the agent (working, napping, needs you, broken);
• a red banner when Hermes itself is unhealthy;
• a clock with ticking seconds, so you can tell at a glance that the screen isn't frozen.

It dims the monitor backlight over DDC/CI when nothing is happening, and brightens it for a minute when something changes. The monitor goes into standby on a daily schedule (e.g. 23:00 to 09:00) and whenever you tell Hermes you're going to sleep or out (/sleep, /awake). Vacation mode (/vacation) keeps it off until you turn it back on. A key press or mouse move on the dashboard machine also wakes it.

Written in Rust with Slint: one binary of about 13 MB that idles at about 0% CPU and about 30 MB RSS. Hermes reports its work over MCP (see MCP.md). PLAN.md has the design; the project board tracks progress.

Commands

hermes-beaver                  # the dashboard (full-screen)
hermes-beaver demo             # made-up data, cycling through scenarios
hermes-beaver demo --scenario busy|idle|overflow|broken|empty
hermes-beaver mcp              # MCP server on stdio (Hermes starts this)
hermes-beaver call <tool> ['<json>']  # one tool call, for scripts (JSON defaults to {})
hermes-beaver display off|on|vacation   # switch the monitor, print one line

Global options: --config <path> (default ~/.config/hermes-beaver/config.toml) and --store <path> (default ~/.local/state/hermes-beaver/events.db).

Development

make test      # unit and integration tests
make lint      # rustfmt + clippy
make demo      # run the demo locally (works under WSLg)

Deploying

Deployment targets and per-host config live in secrets/, which is gitignored:

cp -r secrets.example secrets
$EDITOR secrets/deploy.env secrets/config.toml
make deploy

make deploy builds a release locally, copies it to the host, and runs deploy/install.sh there as root. The script:

• installs the binary, a systemd user unit and an XFCE autostart entry;
• keeps the screen from blanking;
• sets up ddcutil for brightness and xset for standby.

Run it again at any time; it's idempotent. When a release adds MCP tools, send /reload-mcp to Hermes afterwards so the model sees them.

To connect Hermes, give BOBER.md to the agent: it lists the changes to make on the Hermes side.

Configuration

Every key is optional:

timezone = "Europe/Warsaw"   # clock, schedule and `until` times; default: the host's zone

[hermes]
home = "~/.hermes"
agent_jobs = []              # cron job ids whose failures raise the health banner
heartbeat_max_age_secs = 180

[tasks]
stale_after_mins = 30        # Telegram/CLI tasks without updates show "stale?"
finished_rows = 8

[brightness]
method = "auto"              # auto (DDC/CI, else gamma) | ddc | gamma | off
idle = 20
active = 80
active_secs = 60
exit_level = 80

[metrics]
disk = "auto"                # block device behind /

[display]
off_at = "23:00"             # daily standby; both default to none (no schedule)
on_at = "09:00"              # leave it out to stay off until turned on

[ui]
renderer = "software"        # software (~24 MB) | opengl (~95 MB); SLINT_BACKEND overrides