PocketAI Logo PocketAI
Get APK Free
⚡ Local LLM Inference Engine | Zero Cloud. Zero Subscriptions Required.

Your Private AI Assistant.
Running 100% On-Device.

Experience ChatGPT-grade AI conversations directly on your Android phone. No server requests, no telemetry, and zero subscription lock-in. Powered by native llama.cpp and Flutter.

📥 Download APK (Free Release) 🔍 Browse Verified Models
SmolLM2-1.7B-Instruct • Q4_K_M (1.06 GB)
RAM: 1.4 GB / 5.8 GB 19.2 t/s
How do on-device LLMs run completely offline on an Android phone?
Zero Compromises

Built for Performance, Speed & Privacy

PocketAI pairs modern Flutter reactive architecture with low-level native ARM C++ inference.

🔒

100% Offline & Private

Your prompts never leave your handset. Inference and SQLite conversation history reside strictly in your phone's sandbox.

⚡

Native llama.cpp Core

Compiled with 64-bit NEON SIMD optimizations for ARMv8/v9 processors. Delivers sustained 15–25 tokens/second on modern chipsets.

📦

Resumable Model Downloads

Download verified GGUF weights with HTTP Range requests, automatic checksum verification (SHA-256), and safe cache management.

🛠️

Local Tool Calling (Pro)

Empower your assistant with client-side function calling: offline calculator, scratchpad notes, code runner, and structured outputs.

🛡️

RAM Safeguard Engine

Intelligent memory watchdog stops generation before Android triggers OOM kills. Unloads weights safely when switching models.

🔑

Cryptographic License Keys

Zero telemetry verification. Pro licenses are authenticated 100% offline using tamper-proof HMAC signatures.

Device Compatibility

Device RAM & Model Policy

We test every model for thermal stability, memory headroom, and context limits.

Physical RAM Recommended Model Quantization Context Window Expected Performance
4 GB RAM SmolLM2 1.7B Instruct Q4_K_M (1.06 GB) 2,048 tokens ● 14–20 tokens/sec
6 GB RAM Qwen2.5 1.5B / Gemma 3 1B Q4_K_M (1.12 GB) 2,048–4,096 tokens ● 18–26 tokens/sec
8 GB+ RAM Llama 3.2 3B Instruct (Pro) Q4_K_M (1.95 GB) 4,096–8,192 tokens ● 12–18 tokens/sec
Transparent Monetization

Free Forever or Unlock Pro Workspace

Inference runs on your phone, so we don't charge per message. Pro is for power users who want tools and advanced models.

Monthly Billing Annual Billing (Save 33%)

PocketAI Free

$0
Free Forever
  • ✓ Full local offline inference with SmolLM2 1.7B
  • ✓ Unlimited chats & local message history
  • ✓ Model downloader with SHA-256 verification
  • ✓ Zero server telemetry & zero data tracking
  • ✕ Pro models (Llama 3.2 3B, DeepSeek-R1)
  • ✕ Autonomous Tool Calling Workspace
Get Started Free
Most Popular

PocketAI Pro

$9.99
per month via Dodo Payments
  • ✓ Everything in Free
  • ✓ Pro Models: Llama 3.2 3B & DeepSeek-R1 Distill
  • ✓ Autonomous Tool Calling: Offline math, code runner, scratchpad
  • ✓ Extended Context Window: Up to 4,096 tokens
  • ✓ Custom System Personas & prompt templates
  • ✓ Bulk Chat Export (Markdown & JSON)
  • ✓ 100% Ad-Free Guarantee
Instant license key issuance • 14-day money-back guarantee
Frequently Asked Questions

Everything You Need to Know

Got questions about local AI execution, safety, or licensing? We have answers.

How does PocketAI run locally without sending data to servers? +
PocketAI utilizes compiled C++ native binaries of llama.cpp directly on your Android phone's processor. The weights are stored in your application storage directory. When you send a message, the model processes your query in your phone's memory without touching any external API or network endpoint.
Why should I upgrade to PocketAI Pro if models run locally? +
PocketAI Free gives you full access to offline chat with starter models. PocketAI Pro provides an advanced AI power workspace: autonomous client-side Tool Calling (offline calculators, code sandboxing, notes scratchpad), unlocks high-capacity reasoning models (Llama 3.2 3B, DeepSeek-R1 Distill), custom personas, and extended context lengths.
How does recurring subscription activation work? +
When you subscribe via Dodo Payments ($9.99/mo or $79.99/yr), your billing email is automatically activated through our Cloudflare Worker webhook. In the PocketAI Android app, simply go to the Pro tab and enter your email address to unlock all Pro models and autonomous tools. An offline attestation token is cached locally, keeping your Pro workspace active even without internet.
Can I install PocketAI on any Android phone? +
PocketAI requires Android 8.0+ (API 26+) and a 64-bit ARM processor (arm64-v8a). A minimum of 4GB physical RAM is required for stable execution of our starter 1.06GB model.