Poki

Pinky is an open-source desktop application that lets AI agents operate real iPhone and Android phones through MCP while a person watches, approves risky actions, and can take over. It runs on macOS, Windows, and Linux, supports Android phones over USB or Wi-Fi and iPhones running iOS 17.4 or later, and works with agents such as Claude Code, Claude Desktop, Codex, Cursor, and other MCP clients. Its gateway enforces approval rules outside the agent: payments, transfers, messages, installs, deletions, calls, orders, links, and other configured actions can require a click, while banking, wallet, and password-manager apps are blocked by default and agents cannot type into password or one-time-code fields. Pinky provides live mirroring, activity history, screenshots, recordings, HTML reports, phone take-over controls, and an MCP interface over stdio or HTTP. It is free software under the Apache-2.0 license, requires no account, and the repository describes the project as pre-alpha.

ai-phone is a SIP phone daemon for macOS on Apple Silicon that lets an AI agent handle incoming calls or place outgoing calls from configured instructions. It runs speech recognition with Parakeet and speech synthesis with Phonon or Kokoro locally through MLX; the conversation can use a local MLX language model, a Claude Code or Codex subscription, or an OpenAI-compatible endpoint. Phone audio travels over SIP/RTP, while a remote language model receives call text, instructions, and conversation history. After setup, the daemon is controlled with the `ai-phone run` and `ai-phone call` commands. Relay mode exposes JSON-based `say`, `listen`, and `hangup` commands so an external agent can control each turn, and outgoing calls return a transcript when they end. It supports direct SIP addresses and telephone connections through Telnyx, Twilio, SignalWire, or another SIP trunk; the current implementation supports English and one call at a time. The project is distributed under AGPL-3.0-or-later, with bundled PJSIP interface files under GPLv2-or-later.

Pi Pocket is a self-hosted, mobile-first web application for running and supervising Pi coding agents on a user's own machine or phone. It stores model calls, tool calls, subagent activity, sessions, notes, schedules, and other state in a durable database so work can continue after server restarts; the web client receives committed session updates through server-sent events. The app supports live multiplayer sessions with presence, side chat, shared notes, mentions, reactions, pins, invitations, roles, and turn-taking. Users can steer or queue tasks, inspect tool output and diffs, fork or retry sessions, use git branches and worktrees, approve plan-mode changes, schedule messages, run goals until a check passes, and open a Chromium-based browser panel. It also provides artifacts, subagents, push notifications, extensions, codemode, and an optional Lancet Guard that requires approval for selected risky calls. Pi Pocket runs on Linux, macOS, Windows, or Android through Termux, requires Node.js 22.19 or newer, and can expose the app locally, over a LAN, through a Cloudflare quick tunnel, or through Tailscale. It is distributed under the MIT license. The project warns that anyone with steering access can make Pi run commands with the permissions of the host machine, and that Pi Durable is experimental.

ESP32 iPhone Agent is an open-source AI automation tool that lets a vision model control a real iPhone without installing an app, using a computer, AirPlay mirroring, and an ESP32-S3 board acting as a USB mouse and keyboard. UxPlay receives the AirPlay stream, a GStreamer pipeline converts it into JPEG frames, and the computer sends relative pointer, typing, scrolling, and key commands to the board over Wi-Fi. A planner model creates a short plan from the task and current screen, while an actor model examines each subsequent screenshot and selects actions such as tapping, swiping, drawing, typing, pressing keys, going home, or finishing. The server tracks and recalibrates the relative pointer, checks whether the screen changed after each action, and can retry or re-plan after missed actions, repeated actions, dropped links, or invalid model responses. The project provides a Python/FastAPI server and browser-based interface with live phone control, logs, screenshots with action overlays, manual controls, calibration, and an emergency stop. It runs on macOS or Linux with an OpenAI-compatible vision model, either remote or local; the phone requires AssistiveTouch but no Mac, developer account, jailbreak, or installed app. The repository is licensed under MIT.

Port22 is a tool for remotely running and controlling code-focused AI agents. It uses a Mac agent paired with a phone to drive agents such as Claude Code, Codex, OpenCode, and Aider, providing live transcripts, in-phone approval, LAN support, and an end-to-end encrypted relay.