a-ghorbani/pocketpal-ai

★ 6,941TypeScriptAudience · generalComplexity · 2/5Setup · easy

Mindmap

mindmap
  root((PocketPal AI))
    What it does
      Offline AI chat
      No data leaves device
      Model benchmarking
    AI Models
      Phi
      Gemma 2
      Qwen
      Danube
    Features
      Custom Pals
      System prompts
      Performance metrics
    Platforms
      iOS
      Android

mindmap root((PocketPal AI)) What it does Offline AI chat No data leaves device Model benchmarking AI Models Phi Gemma 2 Qwen Danube Features Custom Pals System prompts Performance metrics Platforms iOS Android

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Download a small AI model and have private conversations on your phone without any internet connection.

USE CASE 2

Create custom AI personalities with specific roles, like a cooking assistant or a roleplay character, using the Pals feature.

USE CASE 3

Benchmark how fast different AI models run on your specific phone and compare results on a public leaderboard.

USE CASE 4

Test the same message across multiple AI models to compare their responses side by side.

Tech stack

TypeScriptReact NativeiOSAndroid

Getting it running

Difficulty · easy Time to first run · 5min

Download from App Store or Google Play, no build steps required for end users.

Open source, check the repository for the specific license terms.

In plain English

PocketPal AI is a mobile app that runs AI language models directly on your phone, with no internet connection required. It works on both iOS and Android. All processing happens on the device itself, so your conversations and data never leave your phone or reach any external server. The only information that can exit your device is benchmark data you choose to share or feedback you submit voluntarily. The app lets you download small language models from a built-in catalog that includes options like Phi, Gemma 2, Qwen, and Danube. You pick a model, download it, load it into memory, and then chat with it. You can swap between models at any time, adjust settings like the system prompt and temperature, and retry a message using a different model for comparison. Real-time performance metrics show you how fast the model is responding, measured in tokens per second. A feature called Pals lets you create distinct AI personalities with custom settings. You might build an assistant focused on cooking recipes, or a roleplay character with a specific location and role. The Pals picker in the chat screen makes switching between these personalities quick. The app can also generate a system prompt for a new Pal using AI, so you do not need to write one from scratch. The benchmarking tool measures how fast a given model runs on your specific device and shows memory usage alongside speed metrics. Results can optionally be submitted to a public leaderboard where users compare phone performance across different devices. PocketPal AI is available as a free download on the App Store and Google Play. The repository is open source, and the README includes setup instructions for developers who want to build or contribute to the project.

Copy-paste prompts

Prompt 1

Help me write a system prompt for a PocketPal AI Pal that acts as a personal fitness coach, giving short workout tips and asking about my progress.

Prompt 2

I want to use PocketPal AI with the Gemma 2 model. Write 5 test prompts to evaluate how well it handles creative storytelling.

Prompt 3

Generate a system prompt for a PocketPal AI Pal that acts as a Spanish language tutor, keeping responses short and asking me to repeat phrases.

Prompt 4

I'm comparing Phi vs Qwen in PocketPal AI for coding help. Write 3 sample prompts to test which model explains Python better.

Prompt 5

Write a system prompt for a PocketPal AI Pal that acts as a recipe assistant, suggesting meals based on ingredients I list.

Open on GitHub → Explain another repo

← a-ghorbani on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.