English · Español

A 3D robot on your desktop that runs, watches and approves every AI agent you use — on the plan you already pay for, or 100% local.

APOLO lives at the top of your screen as an animated helmet. It relays Claude Code and Gemini CLI sessions, has its own multi-model agent core (Ollama, OpenAI, Anthropic, Gemini, OpenRouter, Groq…), and asks you before anything risky — from the floating island, a Stream Deck, Discord, Telegram, WhatsApp or your phone.

Beta. Windows 10/11 only for now. Features marked 🧪 are experimental: they work on the author's machine but have had little outside testing.

Why people try it


💳 Use the plan you already pay for	Drive Claude Code and ChatGPT/Codex through their own CLIs — no API key, no per-token bill. APOLO shows usage as % of your plan, never dollars, and pauses its own background work near the limit.
🏠 Local and free	Point it at Ollama and the agent, memory embeddings and routing all run on your PC.
🧩 Universal skills + antivirus	Installs `SKILL.md` skills from a folder, zip or GitHub (`owner/repo`) and also reads the ones in `~/.claude/skills` and `~/.codex/skills`. Every new skill starts disabled and goes through a static + model "antivirus" (green / yellow / red). Any model can use them, not just Claude.
🗳️ Model council	Ask several models the same question (say a local one, Claude and ChatGPT). They answer in parallel, see each other's answers and can correct themselves, and a moderator gives you a verdict with the level of agreement.
🌙 Night shift	Queue jobs before bed. Each one runs in its own git worktree and branch (it never pushes); risky permissions are denied and logged instead of waking you, and at breakfast you get a report plus a ~60 s vertical video of the night.
🎁 APOLO Wrapped	Weekly / monthly cards: hours, streaks, favourite model, tools, what you saved by running locally. Private by default (no message text). Export as PNG or video.
👁️ Printable desk eye 🧪	An ESP32-S3 + round display "eye" (~25 USD) that shows the robot's mood, approves with a tap and denies with a long press. Firmware, wiring guide and simulator included; printable case still in progress. → docs/ojo
🎥 Streaming co-host 🧪	Reads your Twitch / YouTube chat, filters it, reacts in an OBS overlay and talks back. → docs/stream.md
📦 Migrate from OpenClaw	Import memory files (`SOUL.md`, `USER.md`, `MEMORY.md`…), automations, agents and skills from OpenClaw-style assistants in one step.

Desk eye gestures

Everything else

🏝️ Floating island — live sessions, subagents, files being written, plan usage and context per session.
🛡️ Permissions everywhere — Allow / Deny / Always from the island, Stream Deck, Discord, Telegram, WhatsApp, the mobile app or the desk eye. Dangerous commands always need a second confirmation.
🧠 Agent core — tools, persistent semantic memory with nightly consolidation, scheduled tasks and heartbeat, parallel subagents with Mission Control, context compaction, MCP server, web control panel, plugin SDK.
🌐 Browser agent 🧪 — Chrome / Edge / Brave extension with per-site permissions; it never reads or types password or card fields.
🖥️ Computer use 🧪 — sees the screen and drives mouse and keyboard with per-task consent, a red border and a panic stop (just move the mouse).
📱 Mobile app (PWA) 🧪 — pair with a QR code; scoped device tokens; PIN or passkey for dangerous approvals.
🗣️ Voice — neural voice and local Whisper speech input (optional add-on), or plain text.
🔌 Integrations — Discord, Telegram, WhatsApp 🧪, email (Gmail / Outlook / IMAP), GitHub, Hugging Face, ElevenLabs, Stream Deck plugin, Gemini CLI hooks.

Install

Windows installer (recommended)

Download APOLO-Setup-x.y.z.exe from Releases.
Run it. It installs per user (no admin) in %LOCALAPPDATA%\Programs\APOLO, adds Start menu and desktop shortcuts, and asks whether to start with Windows.
The installer is not code-signed yet, so SmartScreen will say "Windows protected your PC" → More info → Run anyway.

Updates arrive on their own from GitHub Releases (differential download); the island asks before installing. Uninstalling removes APOLO's hooks from Claude Code / Gemini CLI and asks whether to delete your data (%APPDATA%\robot-companion).

Silent install: APOLO-Setup-x.y.z.exe /S (add /D=C:\path to choose the folder).

One-line install (PowerShell)

irm https://raw.githubusercontent.com/DMNENGINE/apolo/main/install.ps1 | iex

Installs Node.js, Electron and the voice packages if missing, puts APOLO in %LOCALAPPDATA%\APOLO and follows the main branch. Run it again to update. Both installs share the same data folder.

From source

git clone https://github.com/DMNENGINE/apolo.git
cd apolo
npm install
npm start

Build the installer yourself with npm run dist → dist/APOLO-Setup-<version>.exe.

First steps

From the tray icon:

Install hooks — connects Claude Code (and Gemini CLI if present). It uses node when you have it; otherwise APOLO runs the hook itself.
Open control panel — the welcome wizard picks a model and channels in about two minutes.
Install voice and microphone (optional) — no longer needed: the neural voice and the microphone (whisper.cpp, downloaded automatically, ~200 MB) work without Python. This option installs the Python Whisper in case you prefer it (e.g. with an NVIDIA GPU).
Browser extension (optional) — tray → Open extension folder, then chrome://extensions → Developer mode → Load unpacked, and paste the token from Copy token for the browser extension.
Stream Deck (optional) — run streamdeck\instalar.ps1 from the app folder (resources\app.asar.unpacked\streamdeck in the .exe install).

Supported models

Provider	How	Cost
Claude Code (your Claude plan)	hooks + `claude` CLI	your plan
ChatGPT / Codex (your ChatGPT plan)	`codex` CLI	your plan
Ollama (Qwen, Gemma, Llama… and `-cloud` models)	OpenAI-compatible API	free / local
OpenAI, OpenRouter, Groq or any OpenAI-compatible server	API key	per token
Anthropic API	API key	per token
Google Gemini (API) and Gemini CLI (hooks)	API key / CLI	per token / plan

Models are written as provider/model (e.g. ollama/qwen3.6) and can be set per channel; an automatic router sends coding work to Claude Code and everything else to your default model.

Security

Local API on 127.0.0.1 only, and every request needs a token. LAN access for the phone is opt-in and scoped.
Secrets are encrypted with Windows DPAPI and never shown back; memory refuses to store keys or passwords.
Dangerous actions always ask (twice for destructive ones); one panic (Stream Deck, desk eye, island) denies everything and releases control.
Tamper-evident audit log of approvals and actions.
Plugins run in a sandboxed process; new skills are scanned and stay disabled until you turn them on.
Banking, wallet and password-manager windows and sites are off-limits to screen and browser control.

Details in SECURITY.md. Found a vulnerability? Please report it privately as described there.

Roadmap

One-line installer and Windows .exe installer with auto-updates
Multi-model core, skills engine, model council, night shift, Wrapped
Mobile app, Telegram, WhatsApp, Discord, email
Code-signed installer
macOS and Linux — the core already runs there; desktop control is Windows-only for now (docs/portabilidad.md)
Printable case for the desk eye (STL)
Opt-in anonymous telemetry and crash reports
Skills marketplace in the panel

Full plan: docs/ROADMAP.md (Spanish).

Contributing

Issues and PRs are welcome. Read CONTRIBUTING.md and the Code of Conduct. The core tests need no dependencies (cd core && npm test) and run on Windows and Linux in CI.

apolo