<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Shreyanshk on Medium]]></title>
        <description><![CDATA[Stories by Shreyanshk on Medium]]></description>
        <link>https://medium.com/@shreyansh152k?source=rss-c9580e4ad6ff------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*qVox6eGGAL2IZkRa</url>
            <title>Stories by Shreyanshk on Medium</title>
            <link>https://medium.com/@shreyansh152k?source=rss-c9580e4ad6ff------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 22:32:24 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@shreyansh152k/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Using OpenClaw with Free AI Gateway Models in OmniRoute]]></title>
            <link>https://medium.com/@shreyansh152k/using-openclaw-with-free-ai-gateway-models-in-omniroute-5a9a722b7254?source=rss-c9580e4ad6ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/5a9a722b7254</guid>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[omnirout]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[web-crawler]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Shreyanshk]]></dc:creator>
            <pubDate>Mon, 17 Aug 2026 08:32:48 GMT</pubDate>
            <atom:updated>2026-08-17T08:32:48.574Z</atom:updated>
            <content:encoded><![CDATA[<h4>Building a local AI agent with automatic model routing, WhatsApp, and LAN access</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eIEkvzm4fdLk5vQUjO5Nnw.png" /><figcaption>OmniRoute</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DSIi18R16u3IY03nEoQeSg.png" /><figcaption>OpenClaw</figcaption></figure><p>I wanted to run an AI agent locally while keeping the flexibility to use different AI models without changing the agent configuration every time.</p><p>For this setup, I used:</p><ul><li><strong>OpenClaw</strong> as the AI agent and Gateway</li><li><strong>OmniRoute</strong> as the AI gateway and model-routing layer</li><li><strong>Kiro</strong> as one of the connected providers</li><li>OmniRoute’s <strong>auto/* routing</strong> to automatically select available models</li><li><strong>WhatsApp</strong> as an additional interface</li><li><strong>LAN access</strong> so I can access OpenClaw from devices on the same network</li></ul><p>The final architecture is simple:</p><pre>Phone / WhatsApp<br>       │<br>       ▼<br>   OpenClaw<br>   :18789<br>       │<br>       ▼<br>   OmniRoute<br>   :20128<br>       │<br>       ▼<br> auto/best-free<br>       │<br>       ▼<br>Available AI Models</pre><h3>Why OpenClaw + OmniRoute?</h3><p>OpenClaw handles the agent.</p><p>It manages the workspace, sessions, tools, channels, and interaction with the user.</p><p>OmniRoute handles the model layer.</p><p>Instead of configuring OpenClaw directly with one specific model, OpenClaw can send requests to OmniRoute and let OmniRoute decide which available model should handle them.</p><p>This means I can use:</p><pre>auto/best-free<br>auto/best-coding<br>auto/best-reasoning<br>auto/best-fast<br>auto/best-vision<br>auto/best-chat</pre><p>without changing the underlying provider configuration.</p><h3>1. Install OmniRoute</h3><p>On Windows, install OmniRoute using npm:</p><pre>npm install -g omniroute</pre><p>Check the installation:</p><pre>omniroute --version</pre><p>In my setup:</p><pre>3.8.49</pre><p>Start OmniRoute:</p><pre>omniroute</pre><p>The dashboard is available at:</p><pre>http://localhost:20128</pre><h3>2. Configure the AI Provider in OmniRoute</h3><p>Open the OmniRoute dashboard:</p><pre>http://localhost:20128</pre><p>Go to:</p><p><strong>Providers → Add Provider</strong></p><p>I connected my <strong>Kiro AI</strong> account using OAuth.</p><p>After connecting, OmniRoute automatically displayed the available models from the provider.</p><p>For example:</p><pre>kr/claude-sonnet-5<br>kr/claude-sonnet-4.5<br>kr/claude-haiku-4.5<br>kr/deepseek-3.2<br>kr/qwen3-coder-next<br>kr/gpt-5.6-luna</pre><p>The exact model list depends on the provider and account configuration.</p><h3>3. Use OmniRoute’s Automatic Routing</h3><p>This is the main reason for using OmniRoute.</p><p>Instead of selecting:</p><pre>kr/claude-sonnet-5</pre><p>I am demonstrating with:</p><pre>auto/best-free</pre><p>OmniRoute provides several automatic routing profiles, including:</p><pre>auto/best-free<br>auto/best-coding<br>auto/best-reasoning<br>auto/best-fast<br>auto/best-vision<br>auto/best-chat<br>auto/best-coding-fast<br>auto/pro-coding<br>auto/pro-reasoning<br>auto/pro-vision<br>auto/pro-chat<br>auto/pro-fast<br>auto/coding<br>auto/fast<br>auto/chat<br>auto/claude-opus<br>auto/claude-sonnet</pre><p>For my default setup, I use:</p><pre>auto/best-free</pre><p>This allows OmniRoute to dynamically select from the eligible models available through the connected providers.</p><h3>4. Connect OpenClaw to OmniRoute</h3><p>OpenClaw can communicate with OmniRoute through its OpenAI-compatible API.</p><p>The OmniRoute endpoint is:</p><pre>http://localhost:20128/v1</pre><p>The important part of the OpenClaw configuration is:</p><pre>{<br>  &quot;agents&quot;: {<br>    &quot;defaults&quot;: {<br>      &quot;workspace&quot;: &quot;C:\\Users\\user\\.openclaw\\workspace&quot;,<br>      &quot;model&quot;: {<br>        &quot;primary&quot;: &quot;omniroute/auto/best-free&quot;<br>      }<br>    }<br>  },<br>  &quot;models&quot;: {<br>    &quot;providers&quot;: {<br>      &quot;omniroute&quot;: {<br>        &quot;baseUrl&quot;: &quot;http://localhost:20128/v1&quot;,<br>        &quot;apiKey&quot;: &quot;YOUR_OMNIROUTE_API_KEY&quot;,<br>        &quot;api&quot;: &quot;openai-completions&quot;,<br>        &quot;models&quot;: [<br>          {<br>            &quot;id&quot;: &quot;auto/best-free&quot;,<br>            &quot;name&quot;: &quot;Best Free&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-coding&quot;,<br>            &quot;name&quot;: &quot;Best Coding&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-reasoning&quot;,<br>            &quot;name&quot;: &quot;Best Reasoning&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-fast&quot;,<br>            &quot;name&quot;: &quot;Best Fast&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-vision&quot;,<br>            &quot;name&quot;: &quot;Best Vision&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-chat&quot;,<br>            &quot;name&quot;: &quot;Best Chat&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/best-coding-fast&quot;,<br>            &quot;name&quot;: &quot;Best Coding Fast&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/pro-coding&quot;,<br>            &quot;name&quot;: &quot;Pro Coding&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/pro-reasoning&quot;,<br>            &quot;name&quot;: &quot;Pro Reasoning&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/pro-vision&quot;,<br>            &quot;name&quot;: &quot;Pro Vision&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/pro-chat&quot;,<br>            &quot;name&quot;: &quot;Pro Chat&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/pro-fast&quot;,<br>            &quot;name&quot;: &quot;Pro Fast&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/coding&quot;,<br>            &quot;name&quot;: &quot;Coding&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/fast&quot;,<br>            &quot;name&quot;: &quot;Fast&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/chat&quot;,<br>            &quot;name&quot;: &quot;Chat&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/claude-opus&quot;,<br>            &quot;name&quot;: &quot;Claude Opus&quot;<br>          },<br>          {<br>            &quot;id&quot;: &quot;auto/claude-sonnet&quot;,<br>            &quot;name&quot;: &quot;Claude Sonnet&quot;<br>          }<br>        ]<br>      }<br>    }<br>  }<br>}</pre><p>Replace:</p><pre>YOUR_OMNIROUTE_API_KEY</pre><p>with the API key generated in OmniRoute.</p><p><strong>Never publish your real API key or Gateway token.</strong></p><h3>5. What does openai-completions mean?</h3><p>This does not mean that OpenClaw is using OpenAI as the provider.</p><pre>&quot;api&quot;: &quot;openai-completions&quot;</pre><p>specifies the API format used between OpenClaw and OmniRoute.</p><p>The request flow is:</p><pre>OpenClaw<br>   │<br>   │ OpenAI-compatible API<br>   ▼<br>OmniRoute<br>   │<br>   ▼<br>Selected Provider / Model</pre><p>OmniRoute handles the provider-specific part after receiving the request.</p><h3>6. Restart OpenClaw</h3><p>After changing the configuration, restart the Gateway:</p><pre>openclaw gateway restart</pre><p>Check its status:</p><pre>openclaw status</pre><p>Then open the dashboard:</p><pre>openclaw dashboard</pre><p>The OpenClaw model selector should now contain the OmniRoute models.</p><p>The default model is:</p><pre>omniroute/auto/best-free</pre><h3>7. Switching Routing Profiles</h3><p>The advantage is that I don’t have to change providers manually.</p><p>For example, I can use:</p><pre>auto/best-free</pre><p>for general usage.</p><p>For coding:</p><pre>auto/best-coding</pre><p>For deeper reasoning:</p><pre>auto/best-reasoning</pre><p>For speed:</p><pre>auto/best-fast</pre><p>For vision:</p><pre>auto/best-vision</pre><p>The agent remains OpenClaw.</p><p>Only the routing strategy changes.</p><h3>8. Seeing the Routing in OmniRoute</h3><p>OmniRoute’s <strong>Combo Studio</strong> makes the routing visible.</p><p>For example, selecting:</p><pre>auto/best-free</pre><p>can show the automatic route being resolved into actual provider/model targets.</p><p>In my setup, it showed targets such as:</p><pre>kiro/claude-sonnet-5<br>kiro/gpt-5.6-luna</pre><p>This is useful because it lets you see what OmniRoute is actually doing behind the auto/* model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JxXgVQqFi8qt4EOjxp6N6g.png" /></figure><p><strong>[OmniRoute Combo Studio showing </strong><strong>auto/best-free routing]</strong></p><h3>9. Enable Local Network Access</h3><p>By default, OpenClaw can bind only to the local machine:</p><pre>&quot;bind&quot;: &quot;loopback&quot;</pre><p>That means only the PC can access the Gateway.</p><p>Since I wanted to access OpenClaw from my phone and other devices on the same network, I changed it to:</p><pre>&quot;bind&quot;: &quot;lan&quot;</pre><p>The Gateway section becomes:</p><pre>&quot;gateway&quot;: {<br>  &quot;mode&quot;: &quot;local&quot;,<br>  &quot;auth&quot;: {<br>    &quot;mode&quot;: &quot;token&quot;,<br>    &quot;token&quot;: &quot;YOUR_GATEWAY_TOKEN&quot;<br>  },<br>  &quot;port&quot;: 18789,<br>  &quot;bind&quot;: &quot;lan&quot;<br>}</pre><p>Keep Gateway authentication enabled.</p><p>Then restart:</p><pre>openclaw gateway restart</pre><p>Now the architecture is:</p><pre>Phone<br>  │<br>  │ Same Wi-Fi<br>  ▼<br>Windows PC<br>  │<br>  ▼<br>OpenClaw :18789<br>  │<br>  │ localhost<br>  ▼<br>OmniRoute :20128</pre><p>OmniRoute does not need to be exposed to the LAN.</p><h3>10. Access OpenClaw Using the PC Hostname</h3><p>Instead of relying on the PC’s changing IP address, I can use the Windows hostname if it is resolved by the local network.</p><p>Find the hostname:</p><pre>hostname</pre><p>For example:</p><pre>SHREY-PC</pre><p>Then the OpenClaw address can be:</p><pre>http://SHREY-PC:18789</pre><p>This allows devices on the same network to connect without manually remembering the PC’s IP address.</p><h3>11. WhatsApp Integration</h3><p>OpenClaw can also be connected to WhatsApp.</p><p>Start WhatsApp pairing with:</p><pre>openclaw channels login --channel whatsapp</pre><p>Then scan the QR code from:</p><p><strong>WhatsApp → Settings → Linked Devices → Link a Device</strong></p><p>Once paired, the flow becomes:</p><pre>WhatsApp<br>    ↓<br>OpenClaw<br>    ↓<br>OmniRoute<br>    ↓<br>AI Model<br>    ↓<br>Response</pre><p>This gives me another interface for the same OpenClaw agent without changing the underlying model setup.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lfOvOTxVmqzWP_outaDlJg.png" /></figure><p><strong>[ OpenClaw WhatsApp channel with QR pairing]</strong></p><h3>12. Final Setup</h3><p>The final setup looks like this:</p><pre>WhatsApp<br>                       │<br>                       ▼<br>                ┌─────────────┐<br>                │  OpenClaw   │<br>                │   Gateway   │<br>                │   :18789    │<br>                └──────┬──────┘<br>                       │<br>                  localhost<br>                       │<br>                       ▼<br>                ┌─────────────┐<br>                │ OmniRoute   │<br>                │   :20128    │<br>                └──────┬──────┘<br>                       │<br>                auto/best-free<br>                       │<br>                       ▼<br>             Available AI Models</pre><p>The key configuration is:</p><pre>OpenClaw<br>    ↓<br>omniroute/auto/best-free<br>    ↓<br>OmniRoute<br>    ↓<br>Automatically selected model</pre><h3>Conclusion</h3><p>The main benefit of this setup is that <strong>OpenClaw no longer needs to be tightly coupled to a specific AI model</strong>.</p><p>OpenClaw remains responsible for the agent, tools, sessions, and channels.</p><p>OmniRoute becomes the model gateway responsible for routing requests to the available AI models.</p><p>That gives me a simple local setup where I can use:</p><pre>auto/best-free<br>auto/best-coding<br>auto/best-reasoning<br>auto/best-fast<br>auto/best-vision</pre><p>while keeping the agent itself unchanged.</p><p>For me, the biggest advantage is simple:</p><blockquote><strong><em>I manage the agent once, and let the gateway handle the models.</em></strong></blockquote><h3>Useful Links</h3><p><strong>OpenClaw:</strong><br><a href="https://github.com/openclaw/openclaw">https://github.com/openclaw/openclaw</a></p><p><strong>OpenClaw Documentation:</strong><br><a href="https://docs.openclaw.ai/">https://docs.openclaw.ai/</a></p><p><strong>OmniRoute:</strong><br><a href="https://github.com/diegosouzapw/OmniRoute">https://github.com/diegosouzapw/OmniRoute</a></p><p><strong>OmniRoute Documentation:</strong><br><a href="https://github.com/diegosouzapw/OmniRoute/tree/release/v3.8.49/docs">https://github.com/diegosouzapw/OmniRoute/tree/release/v3.8.49/docs</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5a9a722b7254" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mastering Prompt Design with Vertex AI and Gemini 1.5 Flash]]></title>
            <link>https://medium.com/@shreyansh152k/mastering-prompt-design-with-vertex-ai-and-gemini-1-5-flash-e592ad0bdf47?source=rss-c9580e4ad6ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/e592ad0bdf47</guid>
            <category><![CDATA[google-vertex-ai]]></category>
            <dc:creator><![CDATA[Shreyanshk]]></dc:creator>
            <pubDate>Sat, 19 Apr 2025 12:40:18 GMT</pubDate>
            <atom:updated>2025-04-19T12:40:18.749Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>My Journey Through the Google Cloud Skill Badge Lab</strong></p><p>As Generative AI becomes a powerful tool across industries, one of the most exciting areas to explore is <strong>prompt design</strong> — the art and science of crafting effective instructions for large language and vision models. Recently, I completed the <strong>Prompt Design in Vertex AI</strong> skill badge from <strong>Google Cloud Skills Boost</strong>, and I’m here to share what I learned.</p><h3>📚 The Google Cloud Skills Boost Experience</h3><p>Before diving into Vertex AI Studio, I enrolled in the <strong>Prompt Design in Vertex AI</strong> course offered on <a href="https://www.cloudskillsboost.google/"><strong>Google Cloud Skills Boost</strong></a> — a hands-on learning platform by Google Cloud.</p><p>The course provides:</p><ul><li><strong>Step-by-step labs</strong> that simulate real-world use cases</li><li>Access to the <strong>Gemini 1.5 Flash</strong> model in <strong>Vertex AI Studio</strong></li><li>Guidance on crafting, refining, and testing prompts for both text and image generation</li><li>A digital <strong>Skill Badge</strong> upon completion, which you can share on LinkedIn to showcase your expertise</li></ul><p>What I liked most was how <strong>practical</strong> it felt — no fluff, just you and the tools, solving real problems with GenAI.</p><h3>🛠️ Hands-On with Vertex AI Studio</h3><p>The lab walks you through how to:</p><ul><li>Access and interact with the <strong>Gemini model</strong> inside Vertex AI Studio.</li><li>Use both <strong>text-only</strong> and <strong>multimodal</strong> prompts (images + text).</li><li>Generate creative outputs like <strong>marketing taglines</strong> or <strong>descriptions</strong> from product images.</li></ul><p>One of my favorite exercises was generating <strong>taglines for a sneaker brand</strong> using a product image as input. With just a few refined prompts, the Gemini model returned punchy, brand-worthy taglines!</p><h3>🎯 Key Learnings</h3><ol><li><strong>Prompt Structure Matters:</strong> Clear, contextual, and specific instructions help improve model responses drastically.</li><li><strong>Multimodal Inputs Unlock Creativity:</strong> By combining image + text in the same prompt, Gemini can generate richer, more nuanced content.</li><li><strong>Iterate and Refine:</strong> Prompt design is an iterative process. Small tweaks can make a big difference in the quality of results.</li></ol><h3>🚀 Why This Matters</h3><p>As GenAI tools get more integrated into business workflows, <strong>knowing how to talk to AI models effectively becomes a must-have skill</strong>. Whether you’re in marketing, design, product, or development, prompt design bridges creativity with technology.</p><p>This lab gave me a solid foundation, and I’m excited to apply it to real-world use cases — especially in AI-assisted content generation and customer engagement.</p><h3>✅ Want to Try It?</h3><p>You can take the same course and earn the skill badge here:<br> 👉 <a href="https://www.cloudskillsboost.google/">Prompt Design in Vertex AI — Google Cloud Skills Boost</a></p><p>Or start exploring Gemini models directly in:<br> 👉 Vertex AI Studio</p><p>Thanks for reading! If you’re diving into prompt engineering too, let’s connect and share ideas.<br> Feel free to drop your thoughts or experiences in the comments 👇</p><p>#GoogleCloud #VertexAI #PromptEngineering #GenerativeAI #Gemini #AI #MediumBlog #CloudSkillsBoost #TechLearning</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e592ad0bdf47" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>