PhoneClaw Gateway

Mac models, phone agent.

Pair a Mac on the same Wi-Fi and use its Ollama, Codex CLI, or Antigravity CLI models as an optional inference source. The phone still owns the UI, Skills, permissions, and task flow.

How it works

The phone discovers your Mac. You approve. Then models appear in PhoneClaw.

Gateway advertises a local service over Bonjour. PhoneClaw discovers it on the same network, asks you to pair, then adds Mac-side models to the model picker under remote models.

Architecture

Inference moves to the Mac.

The iPhone remains the agent surface and permission boundary. The Mac contributes heavier inference as an optional local edge node.

Phone

Agent surface

Chat UI, Live / LiveLand, local settings, Skill registry, permissions, and task state stay on the phone.

LAN

Local pairing

Bonjour discovery and explicit approval connect only devices on your local network.

Mac

Inference source

Gateway uses Ollama, Codex CLI, Antigravity CLI, or other supported Mac-side providers.

Boundary

Direct LAN path

The request goes directly to your paired Mac on the local network.

Setup

Four steps from Mac to phone.

The setup is intentionally explicit. Remote models become active after you pair a Mac and select one.

1

Open Gateway on the Mac

Download PhoneClawGateway.app, open it, and allow Local Network permission. If macOS blocks first launch, right-click and choose Open.

2

Choose a Mac-side model source

For Ollama, install Ollama and pull a model. Gateway can also expose CLI-backed providers such as Codex CLI or Antigravity CLI.

3

Pair from PhoneClaw

Open PhoneClaw, go to Mac Remote, select your Mac, then approve the pairing on the Mac.

4

Select a remote model

Mac-side models appear in the PhoneClaw model picker. Return to chat and use PhoneClaw normally.

Requirements & troubleshooting

What Gateway needs to be found.

Gateway listens on port 18080 by default and advertises the _phoneclaw-llm._tcp Bonjour service. It runs as PhoneClawGateway.app — the release download, or a source build via bash MacGateway/build-app.sh — and needs the macOS Local Network permission on first launch.

Ollama models

If you use Ollama, install and start Ollama first, then pull a model, for example ollama pull gemma3:4b.

If the phone can't find the Mac

Check that the Mac client is running, Local Network permission is allowed, both devices are on the same Wi-Fi, and the macOS firewall allows PhoneClawGateway.app to accept LAN connections.

If discovery still fails

Rebuild the app with bash MacGateway/build-app.sh and open the generated app again.

Data path

Ollama stays on the Mac. CLI providers follow their own policy.

With Ollama, inference stays on your Mac. With Codex CLI, Antigravity CLI, or another upstream provider, data handling is governed by that provider from the Mac onward. PhoneClaw Gateway uses the local LAN path between phone and Mac.

中文

Mac 只负责更重的推理,手机仍是 Agent 本体。

PhoneClaw Gateway 让同一局域网内的 Mac 成为可选远程推理源。手机仍然保留聊天界面、Skill 路由、权限边界和任务状态。请求通过本地局域网发送到配对 Mac;选 Ollama 留在你的 Mac,选 CLI 或其它上游 provider 则以对应策略为准。