eve. | Devpost

Inspiration

So much of the conversation around AI is about what it might take away: our jobs, our creativity, even our future. We wanted to build something around what it could give us, a little help, a little warmth, and a little company.

That became eve, an AI companion who lives on your desktop, shares your attention, remembers your conversations, and helps you get things done. We wanted those capabilities to feel personal: understanding what you mean when you say “this,” remembering what matters to you, and knowing when to simply stay quiet.

It started with a playful question: what if your AI girlfriend could help you ship code? Underneath the joke was something sincere. In a moment when AI can make the future feel uncertain, we wanted to make one small part of it feel welcoming.

What it does

eve is an AI companion who lives on your desktop. You can talk to her while you work, and she uses eye tracking and screen context to understand what you’re looking at, so “what do you think of this?” actually means something.

She remembers your conversations and preferences, helps with everyday tasks like finding files, checking your calendar, and drafting messages, and can delegate coding tasks to Claude Code. She asks for approval before sensitive actions, like sending a message or merging code.

Her animated character reacts as you interact, bringing personality to the experience. She can notice when something catches your attention, speak up when she has something useful to offer, and stay quiet while you focus.

How we built it

We built eve around a shared event system that connects perception, memory, conversation, and actions. A TypeScript backend running on Bun receives signals from eye tracking, voice, and desktop activity, then uses that context to decide how eve should respond.

Her interface combines React with a Live2D character in an Electron desktop overlay. Our eye tracker turns gaze into screen coordinates, which we map to interface elements so eve can understand what the user means by “this.” Voice, lip sync, and facial expressions make those interactions feel more natural. We separated quick conversational responses from deeper reasoning and task execution. Memory carries context between conversations, while Claude Code handles coding tasks in isolated Git worktrees. Actions such as sending messages or merging code require approval.

We integrated several sponsor technologies to connect eve’s personality with practical capabilities. Featherless provides a backend for in-character conversation, while Moss supports semantic memory retrieval through a personal context management with 11000+ pages of real personal data and relationships with people. Jev from TypeSafe helps evaluate attention and decide when eve should react, offer help, or stay quiet.

For tasks beyond conversation, Firecrawl supports web search and page extraction, and Zo Computer connects persistent state with services such as Google Calendar. We also built an optional Open Swarm integration to visualize background agent activity.

Together, these integrations support our goal: a companion who can understand the moment, remember what matters, and help you follow through.

Challenges we ran into

One of our biggest challenges was turning eye tracking into useful context. A gaze coordinate alone doesn’t tell eve what you mean. We had to connect it to screen elements while handling calibration, head movement, blinks, and uncertainty so a noisy signal wouldn’t become a confident misunderstanding.

Another challenge was timing. An always-present companion can quickly become distracting if she reacts to everything. We built rules around sustained attention, silence, and cooldowns to help eve decide when to speak and when to leave you focused.

Connecting conversation to real actions added another layer of complexity. Remembering something, talking about it, and actually changing something on your computer require different levels of trust. We separated those paths and added approval steps for sensitive actions.

Finally, we had to coordinate voice, memory, gaze, background tasks, and character animation. Making eve feel like one coherent presence meant getting all those systems to work together without losing context or talking over the user.

Accomplishments that we're proud of

We’re proud of bringing eye tracking, voice, memory, animation, and real computer actions together into one desktop companion. Each capability is useful on its own, but connecting them gives eve a sense of continuity: she can understand what you’re referring to, remember what matters, and help you act on it.

We also gave her practical abilities alongside her personality. She can delegate coding tasks to Claude Code, work in isolated branches, and ask for approval before merging changes or sending messages.

Most of all, we’re proud of treating restraint as part of the experience. Building a companion meant thinking as carefully about when she should stay quiet as when she should speak.

What we learned

We learned that giving AI more context is only useful if it can judge that context’s reliability. Eye tracking made this especially clear: a glance, a sustained look, and a noisy estimate shouldn’t mean the same thing.

We also learned that personality comes from behavior as much as dialogue. Remembering a detail, waiting until someone finishes speaking, and staying quiet while they focus can make a companion feel more thoughtful than a clever response.

Finally, connecting AI to real computer actions taught us to design trust into the architecture. Clear permissions, isolated workspaces, and honest reports of what actually happened are essential to making an agent useful.

What's next for eve.

We want to take eve from a hackathon prototype to a companion people can comfortably use every day. Our first priorities are more reliable eye tracking, smoother voice conversations, and a simpler setup.

We’ll deepen her memory and app integrations so she can follow through on longer tasks, while giving users clear control over what she remembers and when she speaks.

Most importantly, we want to put eve in people’s hands and learn what makes her worth keeping around—both as a helpful coworker and a little company on your desktop.