The prometheus command is installed with the package. In a checkout, the make
targets are aliases for the same things, plus the eval and tracing tools.
| Command | Does |
|---|---|
prometheus |
chat in the terminal |
prometheus dashboard |
the live cockpit at localhost:7777 (+ Telegram if TELEGRAM_BOT_TOKEN is set) |
prometheus voice |
talk to it — the "prometheus prometheus" wake word, or push-to-talk (needs the [voice] extra) |
prometheus telegram |
message it from your phone (needs TELEGRAM_BOT_TOKEN) |
prometheus discord |
answer in a Discord server (needs DISCORD_BOT_TOKEN) |
prometheus whatsapp |
answer WhatsApp messages (needs WHATSAPP_TOKEN and a public URL) |
prometheus brief |
a morning briefing from calendar + mail + memory, run as a loop |
prometheus gather |
the same job as a graph: four sources fetched together, then one digest |
prometheus evals turns |
grade your own traced turns with the five turn checks and list the failing ones (--window today, 7d or all) |
prometheus connections |
every integration and its health, including Prometheus Memory |
prometheus connect google |
sign in to Google Calendar (opens your browser) |
prometheus connect waku-memory |
connect Prometheus Memory, the memory shared with your other agents (opens your browser) |
prometheus mcp |
MCP servers, and which account each one knows you as |
prometheus mcp login <name> |
sign in again — as someone else, or after expiry |
prometheus mcp logout <name> |
forget a server's token |
prometheus skill install <url> |
install a community skill from a GitHub or Gist link |
prometheus skill export --to claude,codex |
copy Prometheus's skills to Claude Code and Codex (--project for ./.claude/skills, --force to replace) |
In the dashboard chat, /connect google and /connect waku-memory do the same
as their prometheus connect commands, and /help lists the graph workflows.
| Command | Does |
|---|---|
make run |
chat with Prometheus in the terminal |
make dashboard |
the dashboard at localhost:7777 (restart it after pulling backend changes) |
make voice |
push-to-talk, or always-on with PROMETHEUS_WAKE_WORD |
make telegram · make discord · make whatsapp |
the messaging gateways |
make brief · make gather |
the morning briefing, as a loop or as a graph |
make eval |
deterministic evals (0/1, no judge involved) |
make eval-judge |
LLM-as-judge evals (scored %, needs an API key) |
make gate |
the release gate: deterministic must pass, judge must clear its threshold |
make lint |
ruff over the code and the evals |
make trace |
trace waterfalls in Phoenix at localhost:6006 |
make shootout RUNS="…" |
the same tasks on different models, e.g. RUNS="kimi:kimi-k3 anthropic:claude-opus-4-8" |
make shootout-coding RUNS="…" |
a coding round through pi, scored by tests |
Tests live in evals/, not tests/. evals.md explains the two
kinds.