The latest articles on DEV Community by Ismail Abdullah (@ismail_abdullah_8feb1151a). https://dev.to/ismail_abdullah_8feb1151a https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3854354%2F9e581e2a-fedd-49d5-bcc3-1018792e7b8f.jpg https://dev.to/ismail_abdullah_8feb1151a en Ismail Abdullah Tue, 31 Mar 2026 23:57:10 +0000 https://dev.to/ismail_abdullah_8feb1151a/codey-v2-is-out-stable-release-2fh1 https://dev.to/ismail_abdullah_8feb1151a/codey-v2-is-out-stable-release-2fh1 <p>A persistent, daemon-based AI coding agent that runs fully on-device (Android via Termux or Linux) with zero cloud dependency by default.</p> <p>Works out of the box with:</p> <ul> <li>Qwen2.5-Coder-7B (main coding/reasoning agent)</li> <li>Qwen2.5-0.5B (planner + summarizer)</li> <li>nomic-embed-text-v1.5 (RAG embeddings)</li> </ul> <p>All served locally via llama.cpp.</p> <p>Super easy OpenRouter setup (cloud fallback or free models):</p> <ol> <li>Get your key at openrouter.ai/keys (starts with <code>sk-or-</code>)</li> <li>Set: </li> </ol> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code> <span class="nb">export </span><span class="nv">OPENROUTER_API_KEY</span><span class="o">=</span><span class="s2">"sk-or-your-key-here"</span> <span class="nb">export </span><span class="nv">CODEY_BACKEND</span><span class="o">=</span><span class="s2">"openrouter"</span> </code></pre> </div> <p>Add to ~/.bashrc and <code>source ~/.bashrc</code></p> <ol> <li>That's it! Defaults to <code>qwen/qwen-2.5-coder-7b-instruct</code> for coding + planner. No local models, no llama-server, no daemon needed.</li> </ol> <p>Optional: Override with <code>OPENROUTER_MODEL</code> and <code>OPENROUTER_PLANNER_MODEL</code> (e.g. free or powerful models from OpenRouter).</p> <p>Just <code>pip install -r requirements.txt</code> and run <code>python main.py "your task"</code>.</p> <p>Highlights:</p> <ul> <li>RAG-powered local knowledge base</li> <li>Recursive self-refinement (draft → critique → refine)</li> <li>Smart error recovery &amp; adaptive tools</li> <li>Git integration (branches, AI commits, conflict resolution)</li> <li>Voice support (TTS/STT on Termux)</li> <li>Thermal &amp; resource awareness (great for mobile)</li> <li>Auto Python linting + /review command</li> <li>Persistent state across sessions</li> <li>Fine-tuning ready (export history for Colab)</li> </ul> <p>Full privacy on-device or quick switch to OpenRouter. One-command local install with <code>./install.sh</code> + <code>codeyd2 start</code>.</p> <p>Check it out and start coding!<br><br> 👉 <a href="https://github.com/Ishabdullah/Codey-v2" rel="noopener noreferrer">https://github.com/Ishabdullah/Codey-v2</a></p> <h1> LocalAI #CodingAgent #OnDeviceAI #OpenRouter #Termux </h1> agents ai coding llm