Kevin (formerly Bob)
Wire an LED, tell Kevin, and he writes and installs the skill himself
Pi 5 4 GB $130 + speakerphone; Gemini API key free tier
Prices: $130.00 at Adafruit (checked )

By arancool3000 (GitHub repository owner)
- GitHub stars
- 2 (GitHub)
Build sheet
- Board
- Raspberry Pi 5
- Other parts
- GPIO peripherals added by the user
- AI model or service
- Voice assistant LLM that generates and installs device skills from speech
- Difficulty
- Intermediate
- Setup time
- About 20–40 minutes: card write 10–20 min, first-boot install 10–20 min
- Accounts needed
- Gemini API key (Google AI Studio), Claude Pro/Max or ChatGPT Plus/Pro (optional)
- License
- MIT
Parts list
- Raspberry Pi 5, 4 GB or 8 GB (recommended) (Pi 5 2 GB fine for Kevin alone; Pi 4 2 GB+ works; Pi 3B+/Zero 2 W not supported) [2]
- Case with fan or heatsink (Pi 5) (The wake word listens all the time) [2]
- Official power supply (27 W USB-C for Pi 5) [1]
- microSD card (16 GB+, A1/A2 rated; plus a USB card reader if your PC has no slot) [1]
- USB speakerphone (Easiest option (echo cancelling); or USB mic + speaker. Pi 5 has no headphone socket) [1]
- Breadboard, jumpers, LEDs + 330 Ω resistors, buttons, relay module, DHT22/BME280 sensors, servos with own supply (For building skills) [optional] [1]
- Arduino Nano, 16×2 I²C LCD, joystick module, optional buzzer (Bedside screen) [optional] [3]
Prices and notes: Parts and requirements
Overview
Open-sourced Pi 5 assistant that starts with no device skills; you wire hardware (e.g. LED on GPIO 17) and tell it what you connected—it writes the skill, reads it back, and installs only after verbal confirmation.
What it does
- Starts with no device skills. You wire hardware and tell him, e.g. "I've connected an LED to GPIO 17. Make a skill that turns it on and off." He writes the skill, reads back what it will do, installs only after you say yes, tests it, and keeps it. [1]
- Talks with Google's Gemini Live voice (eight voices, web search). [1]
- Wake phrase "Hey Kevin" is recognised offline on the Pi by Vosk; nothing leaves the Pi until then. [1]
- Named timers, wake-up alarms that ring until stopped, internet radio and music from a folder. [1]
- Answers from free no-account services: rain next hour, UV, air quality, Wikipedia, definitions, currency, holidays, maths, units, recipes, the ISS, trivia. [1]
- Phone remote (/app) and a night-time StandBy clock (/standby); settings page at kevin.local:8080. [1]
- Updates himself from tagged GitHub releases daily and rolls back if a release fails to start. [1]
Things you can say or do
- "Will it rain in the next hour?" [1]
- "Remember the spare key is in the blue box" [1]
- "Set a pasta timer for 10 minutes" · "Wake me up at 7:30" · "Snooze" [1]
- "Play Radio Paradise" · "Next song" · "Quieter" [1]
- "What's 15% of 240?" · "Define serendipity" [1]
- "I've wired a button to GPIO 27, make a skill that tells me if it's pressed." [1]
- "Roll back the desk LED skill." · "Switch off the desk LED skill." [1]
- First, read-only: "Find what's on I2C", then "Install the example temperature skill" (BME280/BMP280) [1]
Use-case ideas
- Starter wiring projects with ready phrases: LED on GPIO 17, button on GPIO 27, relay module on GPIO 22, servo on GPIO 18, BME280 temperature sensor. [2]
- Bedside screen: an Arduino Nano + 16×2 LCD + joystick shows time, weather and timers, and its button talks to Kevin. [3]
- Hand hard questions to your own Claude or ChatGPT subscription. [4]
- Our suggestionA first electronics kit for a kid or classroom, where the assistant explains and builds each new part.
Parts and requirements
| Part | Spec / notes | Needed | Price |
|---|---|---|---|
| Raspberry Pi 5, 4 GB or 8 GB (recommended) [2] | Pi 5 2 GB fine for Kevin alone; Pi 4 2 GB+ works; Pi 3B+/Zero 2 W not supported | Required | $130.00Adafruit, checked |
| Case with fan or heatsink (Pi 5) [2] | The wake word listens all the time · Price is the official Active Cooler, our pick | Required | Our suggestion$13.50Adafruit, checked |
| Official power supply (27 W USB-C for Pi 5) [1] | Required | $14.04Adafruit, checked | |
| microSD card [1] | 16 GB+, A1/A2 rated; plus a USB card reader if your PC has no slot | Required | |
| USB speakerphone [1] | Easiest option (echo cancelling); or USB mic + speaker. Pi 5 has no headphone socket | Required | |
| Breadboard, jumpers, LEDs + 330 Ω resistors, buttons, relay module, DHT22/BME280 sensors, servos with own supply [1] | For building skills | Optional | |
| Arduino Nano, 16×2 I²C LCD, joystick module, optional buzzer [3] | Bedside screen | Optional |
- RAM: Pi 5 4/8 GB recommended; 2 GB OK; Pi 4 2 GB+ works.
- OS: Raspberry Pi OS Lite (64-bit)
- Network: Ordinary home Wi-Fi (not hidden/enterprise) or Ethernet, with internet
- Computer for setup: Windows or Mac for the setup app, or any computer for the manual install
- Gemini API key (Google AI Studio): required, Free tier with usage limits; Google may use free-tier data [1]
- Claude Pro/Max or ChatGPT Plus/Pro: optional, Paid plan [4]
- Requirements source: [1]
How it works
Vosk listens offline for "Hey Kevin". Then audio streams to Gemini Live, which answers by voice and calls tools: built-ins (time, weather, timers, notes, volume, Pi status, build skills) or your own skills. A new skill is shown with its exact code, a diff and what it can reach on the settings page, and is installed only after a yes that the code enforces.
| Layer | Component | Runs on |
|---|---|---|
| Wake word | Vosk, offline [1] | Pi |
| Voice in/out + tools | Google Gemini Live API [1] | Cloud |
| Skills | One Python file each (gpiozero, smbus2, pyserial, urllib), run by a skill host as user bob-skill, 30 s per call [5] | Pi |
| Hard questions (optional) | Claude Code or Codex CLI, no tools, 2-minute limit [4] | Pi → cloud |
| Settings + phone remote | Web app on port 8080 [1] | Pi |
Setup effort
- Skill level
- Beginner
- Time
- About 20–40 minutes: card write 10–20 min, first-boot install 10–20 min
Low with the setup app: fill in Wi-Fi, passwords and the Gemini key, write the card, boot. Manual: one curl | sudo bash install over SSH. [1]
Limits and caveats
- Needs internet and a Gemini key; conversations go to Google. On the free tier Google may use the data. [6]
- Skills can't watch and speak by themselves ("tell me when the button is pressed" isn't possible) and can't install libraries. [5]
- A skill's reach list is a report, not a sandbox; skills can use the network. [5]
- Can't be interrupted while speaking. [7]
- Setup apps aren't code-signed yet (SmartScreen / Gatekeeper warnings). [7]
- Very new: 2 GitHub stars, v0.6.1, renamed from Bob in October 2026 (internals still say bob-assistant). [1]
Hardware
- Raspberry Pi 5
- GPIO peripherals added by the user
Like this build? Get the weekly builds
Get the best new AI hardware builds every Thursday. One email a week, free.
Built something? Submit a build
Credits and links
- arancool3000: Creator
- License: MIT
Sources
- github.com/arancool3000/kevin-assistant
- github.com/arancool3000/kevin-assistant/blob/main/docs/HARDWARE.md
- github.com/arancool3000/kevin-assistant/blob/main/examples/bedside-screen/README.md
- github.com/arancool3000/kevin-assistant/blob/main/docs/AI-ACCOUNTS.md
- github.com/arancool3000/kevin-assistant/blob/main/docs/SKILLS.md
- github.com/arancool3000/kevin-assistant/blob/main/docs/PRIVACY-SECURITY.md
- github.com/arancool3000/kevin-assistant/blob/main/docs/WINDOWS-SETUP.md
This page is a short summary written by HardwareLab. Build steps, code and photos belong to the creator; follow “View original” for the full project.
Featured? Add this badge
Built this? Paste the badge into your README. It links back to this page.
Markdown
[](https://hardwarelab.dev/projects/kevin-assistant/?utm_source=badge&utm_medium=referral&utm_campaign=launch_2w)HTML
<a href="https://hardwarelab.dev/projects/kevin-assistant/?utm_source=badge&utm_medium=referral&utm_campaign=launch_2w"><img src="https://hardwarelab.dev/badge.svg" alt="Featured on HardwareLab" width="188" height="20"></a>Related builds
All 43 in Voice Assistants & Companions
AImy
Offline vision-enabled voice assistant on a Pi 5 with an LLM accelerator.
- Raspberry Pi 5
- LLM-8850 accelerator
- camera
Advanced23
