Daimon
A local daemon for macOS that gives any MCP-capable AI client eyes, hands, and a face — screen capture, accessibility tree, mouse/keyboard actions, and an overlay — with a built-in security ceiling.
README
Daimon
A local organ that gives any AI eyes, hands, and a face on your Mac.
Daimon is a local daemon for macOS that lets any MCP-capable AI client — Claude Code, Claude Desktop, Cursor, Codex, Copilot CLI, Antigravity, and more — see your screen, act on it, and show you what it's doing. It speaks the standard Model Context Protocol, so it works with any client and is tied to none.
It is an organ, not a driver: pull, not push. Daimon owns no loop and calls no AI — the client connects over MCP and pulls a sense or moves a hand when it wants. Being fully local and open, you can audit exactly what it does with your screen.
⚠️ Public beta. Daimon works end-to-end (installed, signed, driving real apps) but it is young. Use a sensible ceiling, read the security model below, and report issues.
The triad
👁 Perceive — the senses
| Sense | What | Tool |
|---|---|---|
| Vue | screen capture (pixels) | vue_snapshot, vue_displays |
| Touché passif | accessibility tree of a window | touche_tree |
| Touché actif | the element under a point | touche_probe |
Daimon supplies pixels and structure only — it does no vision/OCR itself; the
client looks with its own eyes. Bounded by default for token cost (max_depth,
root, roles, summary, region, …).
✋ Act — the hands
Acting is a separate organ under a ceiling Daimon enforces itself (default L0, hands off):
| Level | Scope | Gate |
|---|---|---|
| L0 READ | nothing | — |
| L1 NONDESTRUCTIVE | scroll, focus, navigate, hover | none |
| L2 INPUT | click (left/right/middle, double, modifiers), type, key, drag | none, unless the target is a point of no return |
| L3 VALIDATION | engaging buttons | human confirmation on any non-return |
| L4 AUTONOMOUS | full autonomy | none — everything traced |
Tools: main_click, main_type, main_key, main_drag, main_hover,
main_press, main_navigate, main_activate (+ L4-gated held-input primitives).
🪞 Show — the face
A premium, click-through, capture-invisible overlay highlights what the agent targets, ripples where it clicks, and emphasises the exact element you confirm at the gate. Never on an action's critical path. Off by default.
Security model
Daimon is built so an AI can act on your machine safely:
- Daimon enforces the ceiling, not the client — any AI plugs in, none is trusted. The AI can never raise its own limit.
- Points of no return (send / delete / pay / drop-on-Trash …) are classified on the observed element (the AI re-probes the real target — a lying agent can't dodge the gate by mislabelling a button) and gated by a native macOS confirmation dialog. Timeout = deny.
- L4 full autonomy is unlocked only by a human typing a phrase out-of-band;
consent is recorded in an append-only, hash-chained ledger.
no-log = no-act. A forged state file can't escalate to L4. - Secrets never leave: secret-role fields (
AXSecureTextField) and declared apps are blanked in Touché and blacked out in Vue, before anything is served. - Kill the process at any time — the physical override always wins.
Install (macOS)
- Download
Daimon-<version>.dmgfrom the latest release. - Open it and drag Daimon to Applications.
- Launch Daimon — the aperture glyph appears in the menu bar (no Dock icon). First run opens the onboarding window.
- Register your AI clients (one click) and grant Screen Recording + Accessibility when guided.
- Restart your AI client. It now has
vue_*,touche_*,main_*,overlay_*.
The menu-bar dropdown lets you set the hands ceiling (L0–L3), toggle the overlay, re-run setup, and quit, any time.
macOS permissions attach to the app that launches Daimon (your terminal / IDE / AI app), not to Daimon.app — the onboarding explains this.
Supported AI clients (auto-detect + one-click register)
Claude Code · Claude Desktop · Cursor · Windsurf · GitHub Copilot CLI · Codex (CLI + Desktop) · Mistral Vibe · Antigravity (Desktop / IDE / CLI).
Registration is idempotent, reversible, and backed up — a malformed client
config is refused, never overwritten. (Codex and Vibe use TOML; Daimon edits a
# DAIMON:START/END marker block in place and leaves the rest of your config
untouched.)
Run from source
pip install -e ".[dev]"
daimon setup # register into detected clients + guide permissions
daimon serve # the MCP stdio server (what clients launch)
CLI: daimon install [--all] | uninstall | status | onboard | setup.
Set the ceiling in ~/Library/Application Support/Daimon/config/motor.yaml
(or via the menu bar). L4: python -m daimon.motor.control engage.
Build the signed DMG
See build/macos/README.md. Requires Xcode CLT, an
Apple Developer ID, and notary credentials. ./build/macos/build_macos.sh
(use --no-sign for a fast local build).
Tests
PYTHONPATH=src python -m pytest -q
The pure core (guard, reversibility, consent, audit, secrets filter, client registration, tray/menu, …) is unit-tested without macOS; the AppKit surfaces are smoke-validated.
License
GNU AGPL-3.0-or-later. © Arborithm. If you run a modified version as a network service, the AGPL requires you to offer your users its source.
Reference & kinship: Omi — perception/action decoupled on macOS via the Accessibility API.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。