Skip to content

Quickstart

Paulus1337 edited this page Jul 28, 2026 · 3 revisions

Quickstart

1. Run it

phoenix

That is genuinely it. First flight scans your machine for API keys in the environment, an old OpenClaw setup, or a local Ollama, probes real models until one answers, and writes the config for you. If it finds everything it needs, it asks nothing at all.

🔥 openphoenix 0.0.1 - first flight

scanning the nest...
  keys found: anthropic (1), openrouter (11)
probing for a live model...
  ✓ anthropic/claude-sonnet-5 (3.0s)

Reborn. ~/.openphoenix/config.toml (mode 600)
  model     anthropic/claude-sonnet-5

Already running OpenClaw? Phoenix notices and asks first:

Found an existing OpenClaw setup at ~/.openclaw/openclaw.json
Migrating would carry over: 13 API key(s), the Telegram bot token, 5 persona file(s).
Starting fresh leaves OpenClaw untouched and sets up new.
Migrate it? [Y/n]

Press Enter to bring it across, or n to start clean. Saying no changes nothing about your OpenClaw install. Details: Migration.

Want to choose everything yourself?

Run phoenix init with nothing detected, or set PHOENIX_WIZARD=1, and you get five short steps. Every one has a sensible default, so holding Enter gets you a working setup.

  1. Where the brain comes from. Claude, GPT, Gemini, OpenRouter, NVIDIA, local Ollama, or any OpenAI-compatible endpoint you paste a URL for. Keys already in your environment are flagged for you.
  2. Which model. Phoenix asks your provider what it can actually run right now and lists them, so you pick from real names instead of guessing. Type any name instead, and add a fallback for when your first choice rate limits.
  3. Chat apps. Pick any of Telegram, WhatsApp, Discord, Slack, Signal, IRC, Matrix, Mattermost and iMessage. Enter for none, all for everything, or a list like 1,3. Nothing is assumed.
  4. Web UI and HTTP API. Optional. Choose the port, and whether it listens on this machine only, every IPv4 interface, or IPv6 too.
  5. Extras. Browser automation, image generation, voice replies, voice transcription, task board, canvas, embeddings recall, dreaming, heartbeat, audit log. All off unless you pick them.

Then one last question about whether shell commands should wait for your approval.

2. Talk

you › summarize this folder
phoenix › ...

Type /help any time to switch models, privacy, thinking effort and more. See Chat-Commands.

3. Put it in your pocket (optional)

Configure a chat app in step 3, then:

phoenix service install

Phoenix now runs in the background, starts at boot, and answers your messages. Photos, voice notes and PDFs work too.

Worried about privacy? Ghost mode leaves nothing behind: Memory.

Clone this wiki locally