A native Mac, iPhone & Apple Watch companion for your AI workflow.
Claude Code · Codex · Grok Build · Grok Bot · Cursor
Follow tasks, review supported requests, and keep account usage in view.
Download for Mac · Get the iPhone app · Get started · 简体中文
Free & open source · No VibeBuddy account · No analytics · Bring your own AI key
Real app, sample tasks: the 1.3.14 demo dashboard. Screenshots illustrate the interface, not live agent results.
Start a refactor in Claude Code, a test run in Codex and a build in Grok Build. Check a Grok Bot result and your remaining Cursor allowance from the same companion. While your Mac stays running and reachable, you can check a question on your phone, follow a task on your wrist, or ask the voice companion for an update.
| See what matters | Keep work moving | Hear the outcome |
|---|---|---|
| One task view. Needs response, Working and Done, with project, model, current activity and available usage data. | Approve and reply. Review a command or bounded diff preview, answer a question, or continue a supported Codex task. | Voice and summaries. Ask for current task status, read a concise completion summary, or enable read-aloud on Mac. |
| Review on iPhone | Open the task details | Glance at your wrist |
|---|---|---|
 |
 |
 Followed tasks Quick answers Completion summaries Quota at a glance |
Captured from the actual 1.3.14 apps in Demo mode. Watch image shows the in-app quota page. Screenshot provenance.
- On Mac: a searchable menu-bar task list grouped by attention, full dashboard, notch Glance and jump back to the originating terminal or app.
- On iPhone: task details, recent dialogue, supported approvals and replies, Live Activity and Dynamic Island counts.
- On Apple Watch: followed tasks, quick answers, completion summaries, task and quota complications, and Smart Stack relevance. A paired iPhone is required; watchOS controls background refresh.
| Feature | What changed |
|---|---|
| Talk to your tasks | 1.3.11 adds GPT-Live 1 with a separately configurable task reasoning model. OpenAI, Gemini, Qwen and Doubao share task-status and supported action tools. |
| Get a useful completion update | Optional AI summaries focus on results, blockers and next steps. Mac read-aloud has its own provider, model and voice, with voice previews. Model settings. |
| Continue the conversation | Reply to supported waits, steer a running Codex turn or continue an existing task. Recent dialogue and delivery receipts help you see what was sent. Task interaction. |
| Keep the Mac close at hand | Search and filter the menu feed, open task details, and use an explicitly opened QR pairing window. Mac companion update. |
| Follow work on your wrist | Watch quick answers, task controls and completion summaries keep followed tasks close at hand. Watch update. |
| Follow Grok Build and Grok Bot | Track Build CLI tasks, account usage and mode-dependent approval requests. Enable read-only Bot observation, summaries of verified ordinary completions and its separate account quota. Grok Build setup · Grok Bot update. |
| Keep Cursor usage in view | Read allowance from an existing Cursor app or Cursor CLI login, with separate Cursor Models and Other Models pools. Cursor CLI support. |
Mac releases and iPhone/Watch App Store updates ship independently. Newer features need an updated client on the device using them. Check the Mac releases and App Store listing for current versions and availability.
VibeBuddy connects to the agents you already run. Available controls follow the actual connection and pending request.
| Connection | Available integration | Where the boundary is |
|---|---|---|
| Claude Code | Lifecycle hooks, task state, usage, permission decisions and answers to supported questions. | Remote responses need the approval hooks. When the native prompt owns the interaction, respond on Mac. |
| Codex CLI / connected app-server | Task state, usage, supported approvals and questions; create, continue, steer and stop tasks through a connected app-server. | Task control requires the connected server to own the task and expose the required turn/request. MCP elicitation is read-only. |
| Codex Desktop | Observe local task progress and completions; jump back to the app. | Desktop may own a separate app-server. Native Desktop approval coverage is incomplete; use the Mac prompt when no answerable request is available. |
| Grok Build | CLI lifecycle hooks, task state, account usage and configured approval gates. | Whether a remote approval resolves the native prompt depends on Grok's permission mode. |
| Grok Bot | Optional read-only task observation, summaries of verified ordinary completions and separate account usage. | Replies and approvals stay in the official app. Question continuations, automated tasks and turns spanning a disconnect are not yet supported. |
| Cursor | Task state for the Agent panel and the Cursor CLI, supported approvals and questions, queued follow-ups for a running turn, continuation of a finished chat through the Cursor CLI, and account usage with separate Cursor Models and Other Models pools. | Task state and remote responses need the Cursor hooks; without them the agent transcript still reports progress. Cursor exposes no way to interrupt a running turn, and no link that opens a specific chat — a jump brings Cursor forward. |
See the Codex integration contract and agent hook setup. Qwen, Kimi, OpenCode and Antigravity coding-agent adapters are experimental community integrations; their verification is separate from Qwen's supported voice-provider integration.
- Install the Mac companion. Download the latest DMG, drag the app into Applications, and launch it. The distributed Mac app requires Apple Silicon and macOS 14+.
- Connect your coding agent. Open Setup in Mac settings and follow the agent-specific instructions. Manual setup is also available.
- Add your iPhone. Install VibeBuddy: Agent Monitor, choose Pair a phone on Mac, then Scan to pair on iPhone. Start on the same trusted local network. iPhone requires iOS 17+; the current Watch companion requires watchOS 26.5+.
- Try your workflow. Run a short task in Claude Code, Codex or Grok Build and follow it until completion. You can also enable Grok Bot observation or connect Cursor usage. Enable voice or summaries separately if you want them.
Just looking? The iPhone connection screen includes See the demo (no Mac needed). Mac works on its own, too. App Store availability varies by region; build from source if needed.
Choose OpenAI, Google Gemini, Alibaba Qwen or Volcengine Doubao, add your own API credential in the app, review the disclosure and explicitly start a call. Ask “Which task needs me?” or give a clear response to an answerable request. Voice conversation, completion summaries and Mac read-aloud are configured independently. Provider charges apply; task monitoring and button approvals need no AI API key.
Live Activity and Dynamic Island counts update while the phone is connected. Closed-app push requires a running Mac, matching APNs signing configuration, a registered phone and notification permission. Public downloads do not yet include turnkey closed-app push setup; see the APNs setup and distribution decision. Actions always require a reachable paired Mac. Focus, notification settings and watchOS scheduling affect what appears.
flowchart LR
A[Claude Code and Grok Build hooks] --> M[VibeBuddy on your Mac]
C[Codex rollouts and connected app-server] <--> M
B[Grok Bot observation] --> M
U[Account usage including Cursor] --> M
M <-->|Paired local connection| P[iPhone]
P <-->|WatchConnectivity| W[Apple Watch]
M -. Optional AI features .-> V[Your chosen AI provider]
P -. Optional voice .-> V
There is no VibeBuddy cloud, account or analytics service. Mac and phone communicate over your network with bearer-token authentication. Optional AI features send the required audio or task text directly to the provider you choose; configured push notifications pass through Apple. Voice credentials are stored in Keychain. Privacy policy.
The Mac app, iPhone app, Watch companion, daemon, shared Swift models and agent hooks are all in this MIT-licensed repository. The integrations and their limits are documented so other developers can inspect, reproduce and improve them.
| Explore | Start here |
|---|---|
| Build and run | Developer setup |
| Shared models and voice adapters | VibeBuddyKit |
| Agent observation and local server | VibeBuddyMacCore |
| Native apps | Mac · iPhone · Watch |
| Protocol checks and design decisions | Codex probes · Architecture decisions |
| Maintenance history | Releases · Merged pull requests |
Contributions are welcome. Reproduce an integration issue, improve a translation, document a device workflow or submit a focused fix. Start with CONTRIBUTING.md. If VibeBuddy helps your workflow, a star or a concrete account of how you use it helps others discover the project.
m5-paper-buddy inspired the hook-and-transcript approach to agent status; open-vibe-island informed the multi-agent model and Mac Glance. VibeBuddy's Swift implementation was written independently. The app icon was developed with the ip-as-logo skill; the in-app cat is drawn in shared Swift code.
MIT © 2026 Xianwei Zhang. An independent community project; not affiliated with or endorsed by Anthropic, OpenAI or Apple.
For access away from your LAN, see Tailscale remote setup.