Features

Everything runs on your device. Here's what you get.

Text Generation

Run any GGUF model. Compatible with Llama, Mistral, Gemma, Phi, Qwen, and more via llama.cpp. No internet, no API keys, no limits.

GGUF llama.cpp Offline

Specs

  • Configurable: temperature, top-k, top-p, min-p, repeat penalty, context length
  • Function calling with grammar-constrained JSON
  • Thinking mode for supported models
  • Per-model configs saved to database
  • 8-15 tokens/sec on flagship phones

Specs

  • Stable Diffusion 1.5 on-device
  • Text-to-image and inpainting
  • Configurable steps, CFG scale, seed, negative prompts, schedulers
  • NPU acceleration on Snapdragon devices

Image Generation

Generate images directly on your phone with Stable Diffusion 1.5. Text-to-image and inpainting, no cloud required. Snapdragon NPU acceleration when available.

SD 1.5 NPU Inpainting

Image Tools

On-device image processing beyond generation. Upscale, segment, and more — all running locally.

Upscaling Ready
Segmentation (MobileSAM) Ready
Depth Estimation Coming soon
Style Transfer Coming soon
LaMa Inpainting Coming soon

RAG System

Document Intelligence

Feed your documents into conversations. The retrieval pipeline finds relevant content so the LLM can answer from your data. Encrypted RAGs with admin and read-only access levels.

Pipeline

FTS4 BM25 Vector Search RRF MMR

Supported Formats

PDF
Word (.doc/.docx)
Excel (.xls/.xlsx)
EPUB
TXT
Chat History
Neuron Packets

Plugin System

7 built-in tools the LLM calls automatically

Web Search

Search and scrape web content

File Manager

List, read, create files

Calculator

Math expressions and unit conversion

Notepad

Save and retrieve notes

Date & Time

Current time, timezone conversion, date math

System Info

RAM, battery, storage, device details

Dev Utils

Hash, encode, format, text transforms

The LLM decides when to call these during conversations. You can enable or disable individual plugins.

AI Memory

After conversations, the AI extracts and stores facts about you. Next time you chat, it remembers. Stale memories fade over time. You control everything — view, edit, delete from the Memory screen.

Mem0-inspired Deduplication Forgetting curve

How it works

  • 1 LLM extracts facts from conversation
  • 2 Jaccard similarity checks for duplicates
  • 3 New facts stored, duplicates merged or skipped
  • 4 Forgetting curve deprioritizes stale memories

Details

  • 10 voices — 5 female, 5 male
  • 5 languages: English, Korean, Spanish, Portuguese, French
  • Adjustable speed and quality
  • Auto-speak option for hands-free use

Text-to-Speech

On-device voice synthesis powered by Supertonic (ONNX Runtime). No network calls — the AI reads responses aloud, right on your phone.

Supertonic ONNX Runtime 10 Voices

Encrypted Storage

All chat data encrypted at rest with AES-256-GCM. Keys stored in Android KeyStore backed by the Trusted Execution Environment. Export everything as a single encrypted backup file.

AES-256-GCM TEE KeyStore

Security layers

  • AES-256-GCM encryption for all chat data
  • Android KeyStore with TEE hardware backing
  • System backup as encrypted .tnbackup (PBKDF2 + AES-256-GCM)
  • Data never leaves the device