Get started

First run

oara setup is the one canonical path. It probes for your inference server, writes the config, and refuses to write one that cannot work.

Four ways to run it

Command	What it does
oara setup	The rich wizard: identity files, gateway, smoke test. The default, and what most people want.
--fast	Probe, write yaml and an env template, done. Three questions. No identity generation, no smoke test.
--noninteractive	Implies `--fast`. Zero questions: first detected server, CLI gateway.
--gateway-only	Add or change a messaging gateway on an install that is already set up.

Prometheus previously shipped two competing init paths with different outputs and no guidance about which to use. They are one subcommand now; the old --setup flag and the prometheus-init script are thin forwards to it.

The fast path, end to end

This is a real run against a llama.cpp server on 8080. Nothing is edited for effect except the tool-registration log, which is noted below.

oara setup --fast --noninteractive

oara setup --fast --noninteractive
┌─ Prometheus setup (fast) ─────────────────────────────────────┐
  Config will be written to ~/.prometheus/prometheus.yaml
└───────────────────────────────────────────────────────────────┘

Probing for local inference servers …

Local inference: 1 server(s) detected:
  • llama.cpp  @ http://localhost:8080        2ms (1 models, first: gemma-3-27b-it-Q4_K_M)
Env template written to ~/.config/prometheus/env

====================================================================
  CONNECT A CLIENT (Beacon)

    Beacon is the desktop cockpit for this daemon (chat, coding
    runs, documents, dashboards). Get it:
      https://oara.ai/beacon

    Address:  <this machine>:8005
              (or this machine's Tailscale / LAN address, port 8005)
    Token:    minted on first daemon start — re-print with
              `oara token show`
====================================================================

Setup complete. Next steps:
  1. Chat now:       prometheus
  2. Always-on:      oara daemon
  Health check anytime:  oara doctor

Two things about that output The address line prints your machine's own hostname, which is correct but not always the one you want — if you reach the box over Tailscale, use that name. And the “Chat now” line still says prometheus rather than oara; both work, and the alias is on its way out.

What it wrote

One config file. It is plain YAML and you are meant to read it.

~/.prometheus/prometheus.yamlexcerpt

model:
  provider: llama_cpp
  base_url: http://localhost:8080
  model: gemma-3-27b-it-Q4_K_M
  grammar_enforcement: true
  max_tool_iterations: 500
context:
  effective_limit: 24000
  compression_trigger: 0.75
  reserved_output: 2000
security:
  permission_mode: default
  workspace_root: ~/.prometheus/workspace
web:
  enabled: true
  api_port: 8005
  ws_port: 8010
  (gateways, deferred tool loading and learning settings omitted — see Configuration)

Plus an env template at ~/.config/prometheus/env. Secrets go there, never in the YAML and never on a command line.

No server found

If nothing is listening, setup tells you exactly what it checked and writes nothing:

no server

Probing for local inference servers …

Local inference: no servers detected on standard ports.
  Checked llama.cpp:8080, Ollama:11434, LM Studio:1234, vLLM:8000.
  Already running a server on another port or machine?
    oara setup --probe-url http://host:port

To run a local model, install one of:

  Ollama (easiest):
    curl -fsSL https://ollama.com/install.sh | sh
    ollama pull qwen3:8b        # or any model that fits your hardware

  llama.cpp (fastest):
    git clone https://github.com/ggerganov/llama.cpp.git
    cd llama.cpp && cmake -B build && cmake --build build -j
    ./build/bin/llama-server -m models/your-model.gguf -c 32768 --port 8080

Then run `oara setup` again — it will detect the server.

No config was written — nothing usable was detected, and Prometheus
refuses to write a config that is known to be broken.

It exits 2 — not zero, so a script notices, and not a crash either. No file is created. Three ways forward, and the output above already gives you the first one:

Start a server

Any of the four on its usual port, then re-run.

Point at a server somewhere else

A non-standard port, another machine, a CI stub. Both API shapes are tried at the URL you give.

--probe-url

oara setup --fast --probe-url http://gpu-box:8080

Use a cloud provider instead

Seven have presets. Each knows its own default environment variable and model.

--provider	Reads	Default model
anthropic	ANTHROPIC_API_KEY	claude-sonnet-4-6
openai	OPENAI_API_KEY	gpt-4o
deepseek	DEEPSEEK_API_KEY	deepseek-v4-flash
kimi	MOONSHOT_API_KEY	kimi-k2.6
glm	ZAI_API_KEY	glm-5.2
mimo	MIMO_API_KEY	mimo-v2.5-pro
qwen	QWEN_API_KEY	qwen3.7-max

cloud

export ANTHROPIC_API_KEY=...
oara setup --provider anthropic

The key is never a command-line argument --provider reads it from your environment or from the env file, and there is no flag to pass it inline. If the variable is not set, setup says so, tells you where the env file is, and writes nothing: “a cloud config without its key is known-broken, and Prometheus refuses to write a config that cannot work.” Use --api-key-env VAR to read from a different variable, and --model to override the preset.

Then check it

oara doctor reads the config you just wrote and tells you what is actually true about this machine.

PreviousInstall NextConnect Beacon