Humanizing Compute Nodes & SEO Fixes for Agentic Web — July 13, 2026

Today, I set out to enhance the humanization of compute nodes and address some critical SEO issues that had been nagging at me. The day was long, clocking in at around nine hours, but I made significant strides with 13 commits—six of which were new features or improvements, while seven were focused on fixing bugs. My focus on creating a more humanized experience in the compute network is crucial for making the agentic web more relatable and easier to navigate.

One of the standout features I shipped today was a complete rewrite of the compute-node path behind the HUMANIZER_PREFER_NODE setting. This rewrite was essential not only for improving the model sizing but also for ensuring that it’s more aware of the Mixture of Experts (MoE) that we’re integrating. The goal here is to make sure that the compute nodes can dynamically adjust based on user interaction and requirements, creating a more seamless experience. However, this endeavor was not without its challenges; it took considerable time to ensure that the model was appropriately sized for different scenarios, and I had to iterate through several versions before landing on a solution that felt right.

In the midst of this, I also tackled some pressing SEO issues. One fix involved escaping the less-than symbol in the JSON-LD emitter, which was a potential XSS vulnerability. This was a critical step to harden my vibe-coder safety checklist. I realized that even minor oversights in the code could lead to significant security flaws, and it’s vital to maintain a rigorous approach to security as I continue to build out the platform. This fix alone consumed a chunk of my time, but it was necessary to ensure that users can trust the platform.

On the compute side, I introduced capacity tiers and a stake-to-earn hook alongside a dispatch model-fit. This is a foundational step for the compute marketplace that I envision, providing users with a clear understanding of their options and how they can engage with the network. I also rolled out a new tool for joining the compute network through a marketplace page that now features hosted node workers and a live network API. This is a significant leap forward in making the compute resources more accessible and user-friendly. However, I ran into issues with redirecting users correctly from the /compute segment, which I fixed by ensuring that the marketplace page serves properly instead of leading users to a map.

The job management features also received a substantial upgrade. I developed a node-job producer/inspector that allows for enqueuing and reading compute jobs for the pull loop. While integrating this, I had to navigate some complexities with job-signing and deferred processing, which took longer than expected due to various edge cases I encountered. Nonetheless, I’m pleased with the outcome, as it lays the groundwork for a more resilient job management system.

Additionally, I made several improvements to the swarm functionalities. I adjusted the keep_alive setting to 30 minutes on Ollama, which should help prevent intermittent helper calls from timing out due to cold starts. I also capped the number of predictions Ollama can make and raised the helper timeout to 120 seconds, as I discovered that the CPU model was exceeding the previous 45-second cap during real tasks. These changes should enhance the responsiveness and reliability of the system, which is crucial for maintaining a smooth user experience.

Reflecting on today’s work, I’m reminded of the unique challenges of building this platform solo, especially without the financial backing of a larger team. Instead of relying on paid tools like Lovable or Cursor, I’m using Claude Code in VS Code. This approach forces me to be resourceful and creative in solving problems. Each commit is a testament to that resilience, and while the road is filled with bugs and setbacks, the progress I make each day is a step toward my vision of a one-man-show company with a billion-dollar valuation. The work is hard, but it’s rewarding to see the pieces coming together, one commit at a time.