# denkr.ai - Local-First AI Companion

> Private local-first mobile AI project focused on long-term, consistent assistant experience.

## Overview

denkr.ai was built by me in my free time as a personal project. The starting point was my fascination with OpenClaw (by Peter) and a simple question: how can this kind of assistant experience become usable not only for technical people, but also for my mom, family, and friends with no technical background.

I wanted to build something that does not feel like a standard chatbot, but like a reliable companion inside the phone: controlled personality, long-term memory, notebooks, skills, tool routing, web search, YouTube search/transcription, session management, and compaction for long-running conversations. It also includes an MCP direction, API/tool-connection possibilities, and multiple LLM providers so behavior can stay useful and stable across different tasks.

Since ChatGPT became public, I have tested almost every major AI chat product. Because of that, my bar for denkr was high: it had to feel more personal, more consistent, and more human than a typical prompt-response interface. Even now, using it daily still surprises me, and that is exactly the product quality I was aiming for.

This is also important: denkr.ai is not a startup pitch and not a monetization project. I built it alone in my free time, it is free to use, and I plan to keep it free. I made it for myself, my family, my friends, and anyone who wants to use it. My 1–2 year goal is to run a strong open-source model directly on the phone and move toward a truly 100% private local agent experience.

## Challenges

- Maintaining assistant behavior consistency across many sessions
- Creating a controlled assistant personality without generic responses
- Building useful long-term memory without context overload
- Balancing local-first data flows with multiple LLM providers

## Solutions

- Session management and compaction techniques for stable long-term usage
- Notebooks, skills, and tool routing for structured assistant tasks
- Web search and provider routing based on request context
- Controlled personality and response logic focused on reliability

## Outcomes

- Android APK available and actively used privately
- Long-term memory, notebooks, and skills validated as core building blocks
- Multiple LLM providers integrated with routing
- Strong learning ground for conversational UX and product logic in AI systems

## Details

- Category: AI Companion
- Role: Product, UX, AI logic & implementation
- Timeline: April 2026
- Status: in-progress
- Technologies: React Native, TypeScript, Local-First Storage, LLM APIs, Tool Routing, Session Compaction