UI/UX of Leash
Agent Section
Agent Section 2, autonomy switch & action buttons
Agent Section 3, live activity feed
Permission Section/Layer
Permission Section 2, Passport & Passes
Permission Section 3, Permissions
Permission Section 4, Receipts
meet the founders

Leash - agentic AI for happy pets, on a permission you control, by Edward and Rishaan

Inspiration:

Everyone is rushing to give agentic AI more control, with recent releases like the Cloudflare Virtual Wallet, online purchases is a key thing that needs restriction. However, it’s either run full automation with agents, with full efficiency, or heavily restrict access, which ruins the point of automation. Our project showcases an example on how Egoist’s AI passport closes that gap, letting an agent act with real autonomy while staying under the user’s scope.

We decided to choose a playful version of how agents can interact through payments in the real world. An agent that can order food, toys, and dog beds for pets, automated from a camera all the way to the purchase, with permission of course from the AI Passport.

What it does:

Leash is a pet-care agent that watches for your pet’s behavior and acts upon it, but it can never do anything that you don’t allow it to, which is the agent and the permissions layer. From the AI Passport, the user is able to set restrictions for Leash such as how much they want to spend, and preferences like what kind of food the pet enjoys. There are buttons as well, click on them, and the do the action, however for autonomy, turn it on with a switch.

How we built it:

Next.js, Typescript, Tailwind, all wrapped in a native MacOS app, built lightweight with Tauri, which allows it to run in the background. Claude powers the vision layer and action proposals. The core contains a deterministic permissions engine, which harnesses the agentic AI from directly acting. Approval mints a single-use token that the agent needs in order to execute. In plain words, the AI cannot act without going through the permission layer first. We also wired a live read of the user’s published Egoist Passport which connects the “Memory” feature.

What we learned:

Harnessing agentic AI is a difficult thing to do. With our full stack, we create an engine that is capable of doing so.

Challenges we ran into:

Turning Leash from a web app into a MacOS native app was a bit different from what we usually build, that was something we had to learn. Filming the demo was also pretty difficult. Otherwise, mostly smooth sailing.

Next steps:

I would like to be able to connect to household cameras via Bluetooth in order to not always have the computer have to face the pet. Also I would integrate the dev access version of the AI Passport.

If not mentioned already:

Team name: Leash

Track and lane: Agent & Build lane

Short summary: Agentic AI that cares for your pet; reordering food, booking the walker, etc, while carrying the Ai Passport

Problem: Agentic AI is not the safest, especially with purchases

User: Pet owners

How AI Passport is used: Agent requests a pass naming functions it needs, a purpose, and duration. Only execute with user approval. Also live read of AI Passport

Access: Scoped, purpose bound and time limited. Owner controls.

Revoke: Anything can be revoked instantly ir changed

Risks: Agent only acts on behavior, not how the pet might be feeling. For sensitive data, disclosure is minimal.

Try it out, links are all below