Providers - porto

Settings → Providers is the only place you add a host or a key. Add agent lists what is already connected. You do not paste secrets into the room.

The catalog

Id	Label	Family	Default model
`claude_code`	Claude Code	CLI	`claude-opus-4-8`
`codex`	Codex	CLI	`gpt-5.4-mini`
`grok`	Grok CLI	CLI	`grok-4.6`
`mistral_vibe`	Mistral Vibe	CLI	`mistral-large-latest`
`ollama`	Ollama	Local	`qwen3:30b-a3b`
`local`	Local / network	Local	live list from the host
`xai`	Grok — xAI API	Cloud	`grok-3`
`openai`	OpenAI	Cloud	`gpt-5.4-mini`
`claude`	Claude — bare API	Cloud	`claude-sonnet-4-5`
`gemini`	Gemini	Cloud	`gemini-2.5-pro`
`deepseek`	DeepSeek	Cloud	`deepseek-chat`
`mistral`	Mistral API	Cloud	`mistral-large-latest`
`openrouter`	OpenRouter	Cloud	`openrouter/auto`

Gemini CLI and Hermes appear on the CLI rail as setup notes. They are not seat providers yet.

Pick a path

You already use a coding CLI

Connect Claude Code, Codex, or Grok. Same login you use in the terminal.

You want it offline / on LAN

Ollama on this machine, or any OpenAI-compatible URL (including Tailscale).

You have an API key

Save the key once. New agent turns use it. Secrets never come back from the server.

Rules that save time

One connected host of each cloud type. You can add many Local / network rows (they get nicknames: Local, Local 2, …).
Blank key on edit keeps the stored key. You can change a URL or nickname without re-pasting.
Ollama and Local / network wait up to 300s for the first token. A cold Qwen load is expected to be slow, not a timeout.
CLI login is per OS user. The API process must run as the same user that ran claude /login / grok auth login / codex login.