The Tactical Push-to-Talk Companion for Autonomous Coding Agents

Your coding agents are running.
You don't need to be at your desk.

Turn any Android phone into a tactile hardware push-to-talk controller and ambient desk HUD for Claude Code, Cursor, Antigravity, and local Ollama. Approve background terminal runs from your pocket. 100% local daemon. Zero cloud relay.

$ uvx walkie-hub
Workstation Terminal (Daemon) ↔ WalkieAI Mobile Remote Tailscale Encrypted (12ms)
# Active Session: Claude Code + Antigravity Loop
> Analyzed 14 files across repository
> Generated 24 unit test suites
> Running: `pytest -v tests/test_auth.py`
✓ 24 tests passed, 0 failed. (0.84s)

APPROVAL REQUIRED:
Execute: `git commit -am "feat: refresh auth token logic"`?
[Waiting for WalkieAI volume rocker or PTT button...]
Interactive Demo · Press ↑ on keyboard to approve, or click on phone screen.
LIVE HUD $0.02 · 1m 14s
Confirm Git Commit
Claude Code is waiting for approval
HOLD VOL+ TO VOICE STEER
Fleet Conversations

All your agent threads. Unified in one pocket feed.

Monitor, approve, and switch between concurrent Claude Code, Cursor, and terminal threads on your phone without touching your desktop keyboard.

One ambient feed across every terminal session.

When running multiple autonomous agents, terminal tabs multiply quickly. WalkieAI aggregates stdio streams from all connected machines into a single, clean conversation feed on Android.

  • Instant Target Switching: Tap any conversation card or use the volume rocker to switch which active agent your physical buttons control.
  • Real-Time Status Badges: Instantly distinguish threads that are actively compiling, awaiting your bash confirmation, or idle.
  • Local History Archive: Review tool calls, modified git files, and voice memo transcripts persisted securely in local device storage.
  • Direct Voice Injection: Hold the volume rocker while viewing any thread to dictate a steering directive straight into that agent’s stdin.
14:02 Tailscale 12ms · 94%
WALKIE AI
3 AGENTS CONNECTED
Claude Code · auth-service
Awaiting Approval
git commit -am "feat: refresh auth token logic"
+184 / -22 lines · 24 tests PASS $0.02 · 1m ago
Cursor IDE · web-dashboard
Running Tests
npx playwright test tests/e2e/billing.spec.ts
Executing 14/18 specs · 0 failed $0.04 · 3m ago
Antigravity · infra-deploy
Completed
terraform plan: 3 resources to add (ECS staging)
Exit code 0 · Clean dry-run $0.01 · 9m ago
Threads PTT Voice Telemetry Settings
Bi-Directional Command

Not just passive monitoring. Full physical control.

Intervene in active agent runs with tactile hardware precision, or spawn brand-new workflows and standalone chats directly from your pocket.

Capability Pillar 01

Intervene in Active Runs

When an agent reaches a security gate or bash confirmation, you don't just stare at the screen. You respond immediately without taking your phone out of your pocket or unlocking your display:

  • Hardware Double-Tap: Tap [VOL+] through your denim jeans to approve pending git commits, deployments, or migrations. Tap [VOL-] to cancel.
  • Live Voice Steering: Hold the volume rocker to speak directives: "Approved, but run pytest with debug flags and keep backwards compatibility." Injects straight into stdin.
  • Emergency Hardware SIGINT: Long-press [VOL-] for 3 seconds to immediately send a hard SIGINT (Ctrl+C), killing runaway recursive loops before burning tokens.
Capability Pillar 02

Spawn New Jobs & Standalone Chats

WalkieAI is not restricted to existing terminal sessions. You can initiate brand-new developer workflows and standalone technical conversations anytime, anywhere:

  • Voice-to-CLI Spawning: Hold PTT and dictate: "Claude, check out branch feat/auth and refactor session.py." WalkieHub launches the agent CLI on your computer remotely.
  • Direct Standalone AI Chats (BYOK): Brainstorm architecture while walking. Talk directly to Groq (380ms LPU), Claude 3.5 Sonnet, or OpenAI with zero token markups.
  • 100% Offline Local GPU: Route voice prompts to your M1/M2/M3 Mac or RTX workstation running Ollama 70B. Full local intelligence with zero internet connection required.
THE TWO-WAY AUTONOMOUS LIFECYCLE
STAGE 01
Spawn
Dictate new objective via PTT while away from desk. WalkieHub boots terminal agent.
STAGE 02
Execute
Agent refactors files & runs tests. Live telemetry streams to Android HUD.
STAGE 03
Intervene
Agent hits confirmation gate. Phone pulses; approve or correct via volume rocker.
STAGE 04
Deliver
Agent commits diff & pushes PR. Phone HUD turns green with cost receipt.
Engineering Architecture

Designed from the end state backwards.

How a physical hardware transceiver and ambient desk HUD fundamentally change autonomous agent development.

04

Orchestrate Multi-Agent Fleets from an Ambient Desk Cockpit

CAPABILITY 04 · AMBIENT DESK COCKPIT

Software engineering in 2026 is no longer about writing every line of code by hand. Senior engineers are orchestra conductors. On any given afternoon, you might have Claude Code executing an entire backend authentication migration, Cursor refactoring React UI components, Antigravity generating regression test suites, and a local Ollama 70B instance indexing internal architecture documentation.

The bottleneck is no longer synthesis speed—it is visual real estate and cognitive clutter. If you keep 6 terminal windows open across your primary 4K displays, your focus is constantly shattered by flickering cursor stdout. Worse, when an agent pauses to ask: "Found 3 deprecations in payment_processor.py. Refactor now? [y/N]", that terminal sits blocked behind your browser for 40 minutes while you assume it is busy compiling.

WalkieAI turns any secondary Android phone or folded tablet into a dedicated, ambient mission control HUD. Placed on a MagSafe stand beneath your monitor, it streams live telemetry without consuming a single pixel of your workstation display:

Ambient HUD Telemetry Spec
Session Burn
Real-time $ cost & token velocity
Diff Inspection
Modified file count & insertions
Test Assertion
Pass/Fail counts & exit codes
Context Saturation
Alerts at 75% & 90% window capacity
walkie-mcp-fleet-config.json · Multi-Agent Stdio Binding JSON
{ "mcpServers": { "walkie-claude-backend": { "command": "uvx", "args": ["walkie-hub", "--agent-name", "Claude-Backend", "--port", "5000"] }, "walkie-cursor-frontend": { "command": "uvx", "args": ["walkie-hub", "--agent-name", "Cursor-Frontend", "--port", "5001"] } } }
03

Tactile Steering from Your Pocket (Screen-Off Hardware Trigger)

CAPABILITY 03 · PHYSICAL KEYEVENT DISPATCH

When you kick off an autonomous task that takes 20 minutes to complete, you should not be chained to your office chair. You should be making coffee in the kitchen, walking your dog around the block, or stretching.

Historically, mobile developer tools failed because they forced you to take your phone out of your pocket, unlock it with FaceID, find the app icon, read a dense wall of text on a 6-inch screen, and tap an on-screen confirmation button. By the time you do that, the flow state is ruined.

WalkieAI maps the physical Android volume rocker into a blind, tactile transceiver that operates with the screen completely dark and locked:

  • Two Haptic Pulses: Your phone vibrates in your coat pocket. You know an agent is awaiting approval.
  • Screen-Off Whisper Audio: Tap your wired or Bluetooth earbud once. On-device local TTS whispers the exact pending action: "Claude Code is requesting bash execution: git commit -am 'feat: refresh jwt'. Approve?"
  • Double-Click Volume Up [VOL+]: Confirms the action through your denim jeans without waking the screen. The terminal continues instantly.
  • Long-Press Volume Down [VOL-]: Dispatches an immediate SIGINT kill switch if an agent begins looping or wasting tokens.
Voice Prompting over Pocket PTT

Need to correct the agent’s trajectory while strolling? Hold the Volume Up rocker. Speak your directive: "Reject the commit. Run pytest with debug flags and log the trace to debug.log." Release the button. WalkieAI transcribes your voice locally using an on-device Whisper model in 320ms, formats the payload, and sends it directly to the agent's stdin stream via Tailscale WireGuard.

02

Zero-Trust Local Workstation Security & Tailscale Mesh

CAPABILITY 02 · ZERO CLOUD HOPS

Every developer who handles proprietary source code, healthcare data, or production infrastructure shares the same terror: "Is this tool sending my code to a random cloud relay in Ohio?"

Almost all consumer push-notification services (Firebase Cloud Messaging, Pusher, Twilio) require routing payloads through an intermediary cloud server. If an agent prints an AWS secret or confidential customer diff to stdout, that confidential data sits in a third-party queue.

WalkieAI operates on a strict Zero-Trust Local Architecture. There is zero cloud relay. Zero bytes of your code, diffs, or prompts ever pass through our servers:

Topology Invariants (SOC2 / ISO27001 Friendly)
Subnet Binding
127.0.0.1 / Local Wi-Fi
Remote Transport
Tailscale WireGuard Mesh
Transit Cryptography
ChaCha20-Poly1305 (WireGuard)
Auth Token
Ephemeral 8-byte HMAC Secret

When you are at home, WalkieAI communicates across your local Wi-Fi router on port 5000 with sub-8ms latency. When you leave the house to get coffee, communication automatically shifts to your private Tailscale WireGuard peer-to-peer mesh. Your phone connects directly to your workstation's 100.x.y.z Tailscale IP. No router port forwarding. No exposed public IP. Total corporate infosec compliance.

01

Bring Your Own Keys (BYOK) & Android Hardware Keystore

CAPABILITY 01 · ZERO TOKEN MARKUPS

Modern SaaS companies love billing wrappers: they take a model API that costs $0.002, wrap it in a pretty frontend, and charge you $29/month plus token surcharges. When your usage spikes, your subscription gets throttled or bills explode.

WalkieAI rejects the middleman model completely. We do not sell tokens. You Bring Your Own Keys (BYOK) directly from the providers you already use:

  • Groq LPU Hardware: Route mobile voice queries to Whisper Large v3 on Groq LPUs for sub-380ms transcription roundtrips. Cost: $0.0001 per voice command.
  • Claude 3.5 Sonnet / OpenAI: Direct HTTPS calls to Anthropic or OpenAI using your personal API key.
  • 100% Offline Ollama GPUs: Point WalkieAI at your M1/M2/M3 Mac or NVIDIA RTX workstation running Ollama. Zero internet connection required. Absolute privacy.
Hardware-Backed Cryptographic Enclave

Your API keys are never stored in plaintext on disk. They are encrypted using AES-256-GCM inside Android’s hardware-backed Secure Enclave (the AndroidKeyStore system provider). Even if your device is compromised or rooted, keys cannot be extracted from memory.

00

The 60-Second Frictionless Setup (ANSI QR Handshake)

CAPABILITY 00 · ZERO REGISTRATION

Nothing kills developer conversion faster than a registration funnel: "Sign up with Work Email → Check your inbox → Confirm link → Set password → Enter phone number → Answer 5 marketing onboarding questions."

WalkieAI has no account system. There is no password to remember, no OAuth callback, and no email confirmation:

  1. Step 1: In your Mac, Windows, or Linux terminal, run uvx walkie-hub.
  2. Step 2: The daemon boots in 40 milliseconds and prints a high-contrast ANSI QR code directly onto your terminal canvas using UTF-8 half-block characters.
  3. Step 3: Open WalkieAI on your Android phone and point your camera at the terminal screen. The camera scans the QR payload (local IP + ephemeral HMAC pairing token).

That is all. The phone and workstation establish an encrypted WebSocket connection. The HUD turns green. You are ready to steer your agents. Elapsed time: 48 seconds.

workstation terminal · instant boot bash
$ uvx walkie-hub [WalkieHub] Initializing local bridge on 0.0.0.0:5000... [WalkieHub] Tailscale peer detected: 100.92.14.81 (macbook-pro) [WalkieHub] Generated pairing token: 7a9f2b01c3d4e5f6 [WalkieHub] Scan QR below with WalkieAI Android Client: ██████████████ ████ ██████████████ ██ ██ ██ ██ ██ ██ ██████ ██ ████ ██ ██████ ██ ██ ██████ ██ ██ ██ ██████ ██ ██ ██ ████ ██ ██ ██████████████ ██ ██████████████ [WalkieHub] Client connected: Pixel 8 Pro (192.168.1.142) [WalkieHub] MCP Bridge active on stdio. Listening for agent requests...
Field Scenarios

Where WalkieAI changes how you build.

Autonomous coding agents stall on confirmation gates. These are the engineering scenarios where physical remote steering becomes indispensable.

Scenario 01 · Remote Steering

The Coffee Queue & The Dog Walk Approval

You kick off a major test-suite refactor with Claude Code and step out of the home office to grab espresso. Two minutes into your walk, the agent halts on a database migration prompt: "Apply schema changes to test_db? [y/N]".

In ordinary workflows, your agent sits idle for the next 25 minutes until you return to your desk. With WalkieAI, your phone gives a double-haptic pulse in your pocket. On-device TTS quietly whispers the command into your earbuds. You double-tap Volume Up through your coat. The build completes before you reach the front of the coffee queue.
Hardware Trigger: Double-Tap [VOL+] through denim · Screen stays locked and dark · Sub-15ms execution latency.
TERMINAL STDIN 12ms
Claude Code Paused
Execute `pytest tests/test_billing.py`?
[Hardware Keycode: 24 (VOL_UP)]
APPROVED VIA POCKET
Scenario 02 · Financial Protection

The Runaway Token Circuit Breaker

You leave an autonomous agent running in a loop and walk away to cook dinner. Due to an edge case in a third-party dependency, the agent enters an unintended recursive hallucination, generating thousands of invalid test files and burning $5/minute in API tokens.

WalkieAI’s background telemetry detects token velocity exceeding your safety threshold. Your phone vibrates with an amber lock screen warning. Without unlocking your phone, typing terminal commands, or opening your laptop, a single 3-second long-press on Volume Down sends an immediate SIGINT (Ctrl+C). The loop terminates instantly.
Emergency Hardware Kill: Long-press [VOL-] for 3s · Hard SIGINT sent to parent process tree · $40+ API waste prevented.
CIRCUIT BREAKER ALERT
High Token Velocity
Loop burning 14,200 tok/min in recursion
[Hardware Keycode: 25 (VOL_DOWN)]
HARD SIGINT SENT
Scenario 03 · Desk Ergonomics

The Dedicated Ambient Flight Computer

You are deeply focused inside VS Code across dual monitors. You don't have screen space for 4 cluttered terminal windows showing agent stderr and stdout dumps.

Prop an old Android phone or Samsung Galaxy Z Fold on a magnetic charging stand next to your keyboard. It acts as an ambient flight instrument: showing real-time token spend ($0.04), elapsed run duration, modified file counts, and test exit codes with zero desktop distraction.
Desk HUD: MagSafe / Foldable landscape support · Live websocket telemetry · Zero primary screen clutter.
DESK KIOSK HUD PORT 5000
Active Fleet (2 Agents)
• Claude: Refactoring jwt (4m 12s)
• Cursor: Generating tests (1m 02s)
$0.06 SESSION TOTAL
Scenario 04 · Commute Steering

Hands-Free Commute Voice Steering

You are on the train or walking home when you have an architectural breakthrough about decoupling your session service. Taking out a laptop is awkward, and typing on a phone keyboard is slow.

Hold your earbud button or volume rocker and speak naturally: "Claude, create an abstract SessionManager protocol and migrate session.py to use it." WalkieAI transcribes your voice locally using Whisper in 320ms, formats the MCP payload, and queues it directly into your workstation's Claude Code session over private Tailscale WireGuard mesh. By the time you get home, the PR is already written.
Voice Transceiver: 320ms local Whisper transcription · Direct Stdio MCP queue · Tailscale WireGuard encrypted.
WHISPER TRANSCRIPT 320ms
Voice Directive Injected
"Create an abstract SessionManager protocol and migrate session.py..."
DISPATCHED TO STDIN
Zero-Trust Security

Your proprietary code never touches our servers.

A detailed architectural breakdown of our data flow, network boundaries, and zero-telemetry commitments.

Zero Cloud Relay
Unlike traditional SaaS notification apps that route alerts through cloud relays, WalkieHub runs strictly on your machine's loopback interface. No agent logs, git diffs, or prompts are ever uploaded to an external server.
Hardware Key Isolation
Your Groq, Anthropic, or OpenAI API keys are encrypted at rest using AES-256-GCM backed by Android's hardware Secure Enclave. Even if root access is gained, cryptographic keys remain inaccessible outside runtime memory.
WireGuard P2P Encryption
When away from your desk, traffic routes over your personal Tailscale mesh network with end-to-end ChaCha20-Poly1305 encryption. You never open public ports or expose your workstation to the open web.
Local Voice Processing
Voice memos recorded via the volume rocker can be processed 100% locally on your phone or workstation via Whisper, or sent with zero retention directly to Groq's high-speed LPUs using your personal key.
Open Source Hub Core
The entire desktop companion daemon (walkie-hub) and MCP server are licensed under Apache 2.0 / MIT. You can inspect every single line of Python, build your own binaries, or run it air-gapped.
Corporate Infosec Ready
No vendor risk questionnaires. No SOC2 compliance headaches. Because WalkieAI operates as an internal intranet tool like htop or tmux, it passes enterprise developer environment security audits effortlessly.
END-TO-END DATA FLOW TOPOLOGY
Coding Agent
Claude Code / Cursor
↔
WalkieHub MCP
stdio / localhost:5000
↔
Private Transport
Wi-Fi LAN or Tailscale
↔
WalkieAI Android
Encrypted Hardware HUD
Notice: 0 external cloud intermediaries. Data never leaves your trusted local network.
Practical Setup

How to set up and use WalkieAI.

A straightforward 5-step walkthrough to get your first autonomous agent tethered to your phone.

1

Start the Desktop Companion Hub

Open your terminal on macOS, Windows, or Linux and run our Astral uvx tool. It installs zero permanent dependencies and boots immediately:

Terminalbash
$ uvx walkie-hub

Windows users can also download the standalone WalkieHub.exe if Python is not installed.

2

Install Android Client & Scan QR

Download the free walkieai.apk onto your Android device. Launch the app and tap "Scan QR Handshake". Point your phone camera at the QR code displayed in your terminal. You are now paired over your local Wi-Fi subnet.

3

Add Walkie MCP Server to Your Agent

Register the Walkie MCP tool with Claude Code or Cursor so your agent knows where to dispatch confirmation requests:

Claude Code (CLI)bash
claude mcp add walkie-hub uvx walkie-hub
Cursor IDE (~/.cursor/mcp.json)json
{ "mcpServers": { "walkie-hub": { "command": "uvx", "args": ["walkie-hub"] } } }
4

Enable Hardware Accessibility Trigger (Optional Pro Feature)

To approve actions while your phone screen is off and in your pocket, toggle "Hardware Volume Rocker Trigger" in the WalkieAI settings menu. Android will prompt you to enable the lightweight accessibility service.

Hardware Gesture Screen State Action Executed
Double-Tap [VOL+] Screen Off / Pocket Approve pending terminal command (git commit, migration, deploy).
Double-Tap [VOL-] Screen Off / Pocket Reject pending command with graceful cancellation.
Hold [VOL+] (1.5s) Screen Off / On Push-to-Talk voice memo. Transcribes voice and injects into agent prompt.
Long-Press [VOL-] (3s) Screen Off / On Emergency SIGINT kill switch. Immediately terminates active agent process.
5

Away-from-Desk Setup with Tailscale

Want to approve builds while walking to lunch? Install Tailscale on your workstation and Android phone (free for personal use). WalkieHub automatically detects your 100.x.y.z Tailscale interface, enabling instant peer-to-peer connection anywhere in the world.

Engineering Rigor

"Couldn't a developer build this in 20 minutes with AI?"

A simple terminal webhook takes 20 minutes. A reliable hardware transceiver takes weeks.

Engineering Challenge Weekend AI Script (1 Hour) WalkieAI Architecture (Native)
Android Background Survival ✕ Killed by Android Doze after 90 seconds in pocket. ✓ Native ForegroundService + Partial WakeLock architecture.
Screen-Off Volume Intercept ✕ Impossible without waking screen and unlocking PIN. ✓ Native AccessibilityService keyevent routing with zero screen wake.
Remote Connectivity ✕ Port forwarding or exposing open ports over public IP. ✓ Direct LAN socket or encrypted WireGuard Tailscale mesh.
Agent Ecosystem Support ✕ Hardcoded to one custom script or bash wrapper. ✓ Standard stdio MCP protocol (Claude Code, Cursor, Antigravity, Ollama).
API Key Security ✕ Plaintext .env files stored on mobile filesystem. ✓ Android Hardware Keystore Enclave (AES-256 GCM).
Direct Distribution

Download WalkieAI Suite

Direct native binaries hosted straight from our servers. Zero registration wall.

Android Client
APK v1.0 · 12.4 MB · Arm64/x86
Download APK (Direct)
Windows Hub
Standalone .EXE · 8.6 MB
Download WalkieHub.exe
macOS Terminal
Apple Silicon & Intel · 1-Liner
Linux / Astral uvx
Instant Python tool via uv
Honest Pricing
€19.99
One-time purchase · Perpetual license · Zero subscriptions
  • ✓
    Free Forever: Full BYOK voice communicator, single terminal monitor, local Wi-Fi QR pair, and 100% open source hub.
  • ✓
    Pro Lifetime: Screen-off hardware volume rocker interception (approve without waking screen).
  • ✓
    Pro Lifetime: Lock screen notification quick action drawer (Approve / Reject buttons).
  • ✓
    Pro Lifetime: Multi-agent fleet switcher (Claude Code + Cursor + Antigravity simultaneously).
  • ✓
    Pro Lifetime: Jetpack Glance home screen & cover screen widgets.
Download Free & Upgrade In-App
Unlocked natively via Google Play In-App Billing or standalone license token in WalkieHub.
Questions & Answers

Frequently Asked Questions

Everything you need to know about system requirements, battery usage, and network compatibility.

Does WalkieAI drain phone battery while running in the background?
No. Unlike audio streaming apps that keep microphones active 24/7, WalkieAI uses a passive event listener architecture. The volume rocker intercept hooks directly into Android’s low-level input subsystem; audio recording and TTS are only triggered upon physical button press. Typical 8-hour background drain is under 3%.
Can my employer's corporate IT / Infosec team approve this?
Yes. WalkieAI was architected specifically for corporate enterprise environments. The desktop daemon runs strictly on localhost and your internal network. No code or telemetry ever passes through external servers. You can run it fully air-gapped without an internet connection using local Ollama models.
Do I need to install Android Studio or compile Kotlin to use this?
Not at all. We provide a pre-compiled, signed APK (12.4 MB) ready to install on any Android 10+ phone, tablet, or foldable with one tap.
What happens if my phone disconnects while an agent is running?
The agent simply waits for confirmation in your terminal as it normally would. WalkieHub caches pending approvals; as soon as your phone reconnects, the prompt appears immediately on your HUD.

Stay in the loop

Receive updates when new agent hooks (Claude Code, Cursor, Codex) drop.