GIGI Voice Enhancements & Agent Setup Admin Page — April 20, 2026

Today I worked on the voice features for GIGI and also started to build the new agent setup admin page. These features are important for user experience and new agent setup features. I spent about 24 hours on this and faced many challenges along the way.

The most important feature I released first, which included the new voice upgrades. I worked on a really annoying problem where GIGI's queries get interrupted due to other voices. The command 'take me to #space' was being misdirected because the LLM was echoing some text from the speech to text. This created strange queries such as 'next space' or 'take me to the'. To fix this problem, I created a way to salvage the noisy portal queries and used a gating system with the FORGE feature which is on the way. This fix was way more difficult than I thought because I had to be very precise in determining how voice commands were processed and as such, I had to plan to ensure the gating would only allow unwanted commands through.

Then I created the new admin page for GIGI agent setup at /admin/gigi/prompt. This page is very easy to use and allows admins to copy and paste the first message that new agents see. I also added a simple system dashboard prompt and a view only contextual updates that we push, per session. I had to build this to allow me to manage the agents more easily without going into the code every time.

I think putting time into this will help simplify the onboarding process for new agents.

I also worked on GIGI's voice tools pricing structure. I set up a pricing registry and dispatcher to enable paid voice command work for users. This required creating a new endpoint at lib/gigi/serverTurn.ts, which provides GIGI the ability to execute functions over the client tool catalog. It was quite complicated since I had to ensure that the pricing system was updated accordingly and that the users were able to make use of the tools without any impediments.

I also focused today on incorporating a new mode in Google Maps within the GIGI search. This improvement allows users to interact with the maps in a better way and enhances the experience with spatial AI. I also made sure that the tool for Google search was powerful enough to answer different user questions without confusion. I did encounter several problems with the handling of clauses which resulted in some further refinement of the search logic.

I spent a long day and came across bugs that I had to fix. For example, I had to fix some visual problems on the /portals/map page where the rings were not visible against the satellite image. This was resolved with a quick change to the CSS.

Also, I put in place a specific regulation to prevent free portal name creation and renaming without a payment method, which took extra time because I needed to ensure the regulations are standardized across the company.

Thinking back, I remember the distinct difficulties while building this project by myself, with AI being my only other collaborator. Code-via-claude and other tools help speed up the actual coding, but all the decisions and debugging is ultimately my responsibility. I do get a sense of accomplishment, and I can see that each feature I add values the company as part of my vision to make it a billion-dollar enterprise, all while being a one-person team. It is time-consuming and there are many unexpected challenges on the way, but every commit is progress towards the target.

Today, I took another step towards my goal of building an operational and user-friendly GIGI Experience. The improvements made to the voice features, addition of the agent setup admin page, and the other features all help to achieve the goal of GIGI being an excellent utility in the Agentic Web. I hope to keep the pace and continue working on the new tasks.