Startrz

AI assistants got smart… but they still stop where the real work begins. They answer questions. They summarize pages. They explain what to do. But when you ask them to research something, fill a form, compare options, prepare a document, or help you work across your computer, most of them hand you instructions and leave the execution to you. CIARA changes that. CIARA — Control Intelligence Assistant for Real-time Automation — is a voice-first AI desktop companion that turns your computer into a conversational workspace. It lives as a transparent glass-pill overlay on your screen, letting you control your desktop through voice or text without breaking your workflow. Say “Hey CIARA” or press ⌘⇧Space, speak naturally, and CIARA begins turning your request into action. Built with Cursor, Electron, and Python, and powered by ElevenLabs voice technology, CIARA combines LLM reasoning, human-like voice interaction, desktop automation, browser control, and a local-first backend into one real-time command layer for your computer. The core system is built around a SPAV architecture: Sense → Plan → Act → Verify. Before acting, CIARA reads the current screen, understands the user’s goal, breaks the request into achievable milestones, performs each step, and verifies progress before continuing. This makes complex workflows feel simple and natural. Ask CIARA to research a topic, summarize findings, help with homework, navigate a website, fill a form, compare options, or prepare a document — and it does not just tell you how. It starts moving through the task. For example, instead of saying “here’s how to buy a train ticket,” CIARA can begin the actual workflow: search routes, compare schedules, choose options, fill the form, and pause for user approval before sensitive steps like payment. CIARA is not just a chatbot in a floating window. It is a real-time automation layer for the desktop. The interface is built around a glass-pill system that morphs between four states: Idle — a compact 220px pill with mic icon and “Hey CIARA” Listening — a 440px live transcription bar Thinking — a minimal 140px bouncing-dot state Doing — a 320px action state with spinner, app icon, and current task CIARA also supports rich multi-modal responses, including cards, tables, markdown, KaTeX math, code blocks, image viewers, timelines, streaming response cards, and step-by-step plan previews. For larger actions, CIARA can show a plan preview modal with each milestone and ask the user to proceed before execution. It also includes a command panel using ⌥Space for typed prompts, first-launch onboarding, keyboard shortcut setup, system checks, and provider configuration. For web workflows, CIARA connects to a Chrome extension that bridges browser actions like searching, filling forms, extracting data, and navigating pages. This allows CIARA to move beyond text responses and actually help complete tasks inside the browser. CIARA is also local-first. Its Python backend and user data stay on the machine under CIARA_DATA_DIR / ~/.ciara, while users can configure their own API keys for LLM and TTS providers as needed. The goal is not to build another AI assistant. The goal is to make the computer itself conversational. CIARA turns voice into action, making desktop workflows faster, more accessible, and usable for people who are busy, multitasking, or unable to rely on a keyboard. No endless prompting. No copy-pasting instructions. No keyboard required. Just speak, and watch your computer work. Built with Cursor. Powered by ElevenLabs. Designed for a future where you don’t type commands — you just ask, and the computer moves. CIARA is what happens when AI stops yapping and starts using the computer for you.

Submitted 14 May 2026

Tatakae Lodge is an anime-inspired strategic deck-building combat game where poker hands become cinematic battle techniques. Instead of simply playing cards, players build hands like pairs, flushes, full houses, and royal flushes to trigger attacks, combos, limit breaks, and ultimate arts. A weak hand might become a basic strike, while a powerful hand can become a dramatic finishing move like Water Slice, Hinokami Kagura, or Sun Halo. The game solves the problem of making card combat feel more emotional, readable, and exciting. Traditional deck-builders can feel abstract, so Tatakae Lodge turns every move into a visual and audio spectacle: the player understands the strategy of the cards, but feels the impact like an anime fight. I built the project using Zed as my fast AI-assisted development environment, helping me prototype UI, combat flow, and game systems quickly. ElevenLabs is used to bring the game to life with dynamic battle narration, character voice reactions, attack callouts, sound effects, and cinematic audio moments. Every major action can be voiced or enhanced with adaptive sound, making each combo feel like a true fighting-game moment. At its core, Tatakae Lodge asks: What if poker hands were not just scores but superpowers?

Dream Studio is an AI-powered game engine built to make worldbuilding, character creation, and gameplay iteration faster and more accessible. I built it with Kiro’s spec-driven development workflow, using structured requirements, design, and task documents to guide implementation of the world editor, animation editor, runtime packages, and orchestration system. Kiro’s specs system is designed exactly for this kind of structured, trackable feature development. I used ElevenLabs APIs to add voice and audio intelligence to Dream Studio from narration and character dialogue to voice-driven creation workflows and audio-enhanced interactive worlds. ElevenLabs provides the core capabilities needed for this, including text-to-speech, speech-to-text, conversational agents, sound effects, and music generation. The result is a creative engine that turns ideas into playable worlds with less friction, while directly matching the challenge brief to build an AI-powered app using Kiro + ElevenLabs

Trem is an autonomous, asynchronous AI video editing agent that turns raw footage into creator-ready edits like Auto Captions, Motion Tracking, Invisible Jump Cuts or full podcast edit. Built with Cloudflare and ElevenLabs, Trem lets users choose an editing technique, then handles the heavy lifting in the background, analyzing footage, with voice-guided creative direction, and assembling structured, story-driven results faster. Instead of manually piecing clips together, creators interact with Trem like a smart editing partner that understands both the footage and the final format.

CAP is a real-time claim-checking agent built to verify suspicious internet claims before people fall for lies, fake promos, or manipulated “receipts.” For our demo, we focused on a simple but highly shareable scenario: a fake message claiming someone had won 11 million ElevenLabs credits but needed to send money first to redeem it. Instead of arguing or guessing, the user asks CAP. CAP searches for live evidence, checks whether the claim matches anything real, and returns a clear verdict. We built CAP with Firecrawl Search and ElevenAgents. Firecrawl Search gives CAP real-time access to web information and structured content, so it can search beyond static knowledge and evaluate whether a claim is supported by credible sources. ElevenAgents makes CAP feel like a fast, conversational agent that can respond naturally and turn claim-checking into an actual user experience rather than a boring search flow. What makes CAP special is the framing: we wanted fact-checking to feel viral, entertaining, and culturally native to how people actually encounter misinformation today in chats, fake promos, screenshots, and random messages that look just believable enough to trick people. Instead of building a dry verification tool, we built something that feels like a social weapon against cap. Don’t argue. Just ask CAP.