Context Window Mapping
Dynamic Header Parsing
GGUF Binary Header Read: 8192 Tokens
Auto Context Reloading Active [Zero Truncation Crash]
Private On-Device AI for Android
Offline AI that runs entirely on your phone.
Chat, search your documents with local OCR, write inside your keyboard, rewrite notes, translate, and turn your device into a private LAN AI server. Powered by dynamic GGUF context window mapping that automatically scales RAM context without crashing.
Dynamic Context Window Mapping
On-Device RAG & OCR
LAN AI Server
Single Shared RAM Model
Biometric Security
Core Product Modules
One Android app powering complete on-device AI workflows.
Krexel is not just a chat screen. It is an OS-integrated private AI toolkit featuring breakthrough context window mapping, writing tools, document search, note-taking, translation, and local network AI serving.
Dynamic Context Window Mapping
Industry-first architecture for offline Android AI: Krexel directly parses GGUF binary headers to read native model context limits (up to 32k+ tokens) with automated context reloading and fallback handling for LiteRT models, preventing token clipping or OOM crashes.
Document Intelligence & OCR Search
On-device file indexing using visual OCR and PDF text extraction. Ask your phone where a document or detail is, and get grounded answers with local previews without cloud vector DBs.
Shareable Local AI Server Endpoint
Turn your phone into a private local AI server over your Wi-Fi network (Port 11180). Power desktop apps, local scripts, or smart tools with API key authentication and rate limiting.
Keyboard AI
Embedded system keyboard integration with fast zero-latency baseline predictions plus optional AI-boosted rephrasing, 100% offline with zero telemetry.
Notes AI
Embedded note-taking rewriter. Rephrase paragraphs, adjust tone, format lists, or summarize drafts directly inside your markdown notes.
Dual-Engine Architecture
Powered by LiteRT for fast task-specific models and llama.cpp for running state-of-the-art quantized GGUF models on mobile hardware.
Interactive Visual Walkthrough
See Krexel's On-Device Intelligence in Action
Experience real offline document search and dynamic context window mapping directly on your device.
Proof Point 1
On-Device Document Search & OCR
User: "Find tax filing reference in my scanned receipt."
Krexel Indexer: Scanning local FTS5 database...
Local Grounded Answer: "Ref #TX-9021 found in Receipt_2026.pdf (Page 1) [Biometric Hardware Verification Active]"
Proof Point 2
Dynamic GGUF Header Context Reloading
Binary Header Inspector: `GgufMetadataReader` extracted limit = 8,192 tokens
Context Token Calculator: Active RAM Allocated = 8,192 Tokens
Status: Prompt Context Safe (0 Truncation, 0 OOM Crashes)
Unified Memory Architecture
One Loaded Model Powers All App Features
Because Chat, Keyboard AI, Notes AI, Translation, and File Search all share a single loaded model in memory with dynamic context mapping, your device operates with zero model-swapping latency and optimized battery consumption.
Offline + Biometric Security
Your data stays locked on your phone.
Krexel is engineered around local execution. All file indexes, OCR scans, notes, and messages remain strictly on-device, protected by biometric hardware verification and Android Keystore encryption.
- Biometric security gate (fingerprint/face unlock) for sensitive extracted data.
- Keystore-backed AES-256-GCM encryption for stored index credentials.
- Fully functional offline in airplane mode with zero cloud logging or telemetry.
On-Device Controls
What users manage locally
Document Indexing
Indexed OCR text and PDFs stay in local storage, isolated from cloud backups.
Context Window Control
Header-parsed GGUF context length mapping prevents token overflow crashes.
Biometric Vault
Sensitive extracted fields require fingerprint or biometric unlock to display.
Model Management
Load, swap, or offload GGUF models on-demand to optimize phone RAM usage.
Model Support
Bring your own model, or download one in-app.
Krexel includes curated downloadable mobile models and a local file import path for open-weights GGUF models with automatic binary header context mapping.
Curated starters
High-performance edge models including Gemma 4 E2B, Liquid LFM2.5, Qwen 3.5, and MiniCPM5.
Hugging Face discovery
Search and inspect mobile-compatible GGUF models directly within the app before importing.
Phone-aware RAM management
Load models on-demand and offload dynamically based on system memory policies.
Product Roadmap
Future Directions for On-Device AI
Krexel is continuously advancing offline mobile intelligence. Here is where the platform is heading next.
Expanded Agent Actions
Safe, local integrations with core Android capabilities like Maps, Contacts, and SMS to fulfill multi-step tasks.
Multi-Document Reasoning
Cross-referencing and synthesizing insights across multiple offline PDF documents simultaneously.
Context-Window Visualization
Real-time visual memory meters providing users precise visibility over token and RAM allocations.
Next-Gen Mobile NPUs
Enhanced hardware acceleration for emerging open-weights edge LLM architectures.
Zero Cost, Maximum Privacy
Core offline AI capability, completely free.
Krexel runs local models on your phone hardware. There are no subscriptions or cloud server costs because inference runs on your own hardware.
100% Offline
Private
- No cloud data logging
- On-device Document OCR Search
- Biometric security gate
Unlimited
Free Forever
- No subscriptions or paywalls
- Unlimited offline prompts
- All core modules unlocked
Control
Local API
- LAN Server for desktop tools
- API key authentication
- Bring your own GGUF models
| Capability | Standard Cloud AI | Krexel (Local On-Device) |
|---|---|---|
| Internet Required | Yes (Always) | No (100% Offline) |
| Context Limits & Reloading | Fixed Server Limits | Dynamic GGUF Header Context Mapping |
| Document Search (RAG) | Cloud Vector DB Uploads | On-Device Local Index & OCR |
| Shareable LAN Server | Paid Cloud API Usage | Free Local API Endpoint |
| Biometric Protection | None / Account Login | Hardware Biometric Vault |
| Subscription Fee | $20+ / month | Free Forever (Local Hardware) |
FAQ
Frequently Asked Questions
Is Krexel offline?
Krexel is built around on-device processing on Android. Its core AI experience is designed to run locally on your phone once a model is available in the app.
Does it work without internet?
Yes, for core on-device inference after your models are already on the phone. Internet is only required if you choose to browse or download model files from online repositories like Hugging Face.
What engines does it use?
Krexel features a dual-engine architecture: LiteRT (formerly TensorFlow Lite) for high-performance, task-specific models and llama.cpp for running state-of-the-art quantized GGUF LLMs locally on mobile CPU/GPU.
How does Krexel handle model context window limits on Android?
Krexel features dynamic context window mapping—directly parsing GGUF binary headers to read full context limits and providing automated context reloading and fallback handling for LiteRT models to prevent token clipping or memory crashes.
Does Krexel send my scanned documents or photos to the cloud?
No. Krexel's Document Intelligence processes, extracts, and indexes all file text and visual OCR data locally on your device. Nothing ever leaves your phone or gets uploaded to remote vector databases.
How does biometric security protect my extracted file data?
Sensitive extracted information (such as identity tags, credentials, or private document numbers) is encrypted locally using Android Keystore-backed AES-256-GCM storage and requires biometric verification (fingerprint or face unlock) before it can be displayed.
How does the Shareable Endpoint allow my laptop to use my phone's AI?
Krexel can act as a local API server over your private Wi-Fi network. Desktop tools or local scripts can send inference requests directly to your phone's loaded model, protected by local API key authentication and rate limiting.
Can I bring my own model?
Yes. Krexel supports curated downloadable models and local GGUF model imports, allowing you to bring custom open-weights models.
Is Krexel really free?
Yes. Krexel's core on-device AI functionality is completely free. There are no subscriptions, paywalls, or feature locks, because processing is powered by your own device hardware.
Coming Soon
Krexel is getting ready for launch on Android.
The product site is live, and the Android release download link will be added here when Krexel launches publicly.