Build Journal
Map Fixes, GIGI Enhancements & Embed Improvements — May 10, 2026
I tackled significant fixes for the map, GIGI, and embed functions, enhancing the user experience on the agentic web platform.
What shipped
- GIGI orb lunar mode fix — GIGI orb now correctly caps at zoom level 5 for lunar mode.
- Improved map stability — Fixed freeze issue during slide-over interactions.
- Scanner organization — Default-All split into #Names and Live Now sections.
- GIGI deduplication — Silenced duplicate ask_gigi responses for clarity.
- Embed enhancements — Email portal owner on every captured lead for better engagement.
Today was a marathon 16-hour day focused on fixing critical issues with the map, GIGI, and embed functionalities of my platform. I set out to enhance user experience and ensure seamless interaction with the agentic web. What I shipped today includes eight important fixes that tackle various bugs, all aimed at refining the way users engage with the spatial AI portal.
The first fix addressed an issue with the GIGI orb in lunar mode, which was bypassing the Earth zoom gate. This was a significant problem since the lunar view should cap at zoom level 5. I spent a good chunk of the morning debugging the zoom functionality, trying to replicate the issue before finally isolating the root cause. It turned out to be a misconfiguration in the zoom gate logic that allowed the orb to exceed its intended limits. After a few iterations, I got it fixed, ensuring that users can now enjoy a proper lunar experience without straying into Earth territory.
Next, I tackled a freeze issue caused by the ?dist=/?scope= URL writes when the slide-over was open. This bug was particularly frustrating because it disrupted the user experience, causing the wheel-zoom feature to rerun searchParams effects and snap the camera back to unwanted positions. After combing through the code for a few hours, I implemented a solution that prevents the searchParams from being affected while the slide-over is active. This fix not only improves usability but also reinforces the stability of the map interactions.
I also split the scanner default-All into #Names and Live Now sections. This change enhances clarity for users by organizing the data better. Additionally, I gated the GIGI orb to zoom levels greater than or equal to 7, which corresponds to roughly a 10km radius. This helps ensure that GIGI is only accessible when users are at an appropriate zoom level, keeping the interface clean and intuitive.
On the GIGI front, I silenced the GIGI response for duplicate ask_gigi invocations. Previously, if a user asked GIGI the same question multiple times in quick succession, they would receive cached replies, which could lead to confusion. Now, if a duplicate request is made within an 8-second window, GIGI simply returns an empty response. I also added a deduplication layer for reply text and adjusted the scanner network-error retry to three attempts, softening the experience in case of transient connectivity issues. These changes should make interacting with GIGI feel smoother and more responsive.
The embed functionality also received significant attention. I fixed an issue where the name anchor caused a description-as-identity hallucination. This was a critical fix as it prevented users from receiving misleading information. Additionally, I implemented a feature that emails the portal owner on every captured lead, which is vital for maintaining engagement and follow-up opportunities. I also ensured that the greeting uses the owner's cloned voice when active, filtering out SEO-spam noise from knowledge retrieval, and pinned the homepage chunk to the current context, enhancing contextual relevance.
Lastly, I enabled Range support and CORS on the /api/files endpoint. This was necessary for iOS Safari to play cross-origin TTS audio for greetings and replies. Ensuring compatibility across different platforms is essential, especially as more users access the agentic web from mobile devices. This fix was a bit tricky, involving some trial and error with CORS headers, but ultimately it worked out well.
Reflecting on the day, it’s clear that building this platform as a one-man show with AI as my team is no small feat. Fixing these bugs required deep dives into the code and a lot of patience, but seeing the progress is incredibly rewarding. Each of these fixes brings me one step closer to my goal of creating a billion-dollar valuation company. As I continue to refine the features and improve stability, I can feel the potential of the agentic web becoming more tangible.
Overall, today’s fixes have significantly enhanced the functionality of the map, GIGI, and embed systems. I’m proud of the work done today, even if it came at the cost of a long day and some frustrating bugs. I’ll carry this momentum into tomorrow as I continue to build and improve the platform. The journey of a solo builder is challenging, but moments like these make it worthwhile.