Sep 29, 2026 4 min read Showcase

AI agents shipped in 12 hour long Invide AI Agent Buildathon

On this Sunday, developers got together on invide Discord to turn their ideas into working projects across agent reliability, security, data workflows, radio, and local AI assistant.

120 people applied to join invide's AI Agent Buildathon. The builders who took part had 12 hours to turn their idea into something they could show. They could work solo or with a team, choose their own technology, but have a single focused goal to finish the project by the end of the day.

Together, they explored what an AI agent can do when given a clear job: remember a failure, help a security team spot unusual behaviour, review a data pipeline, assemble an audio news update, or act as a private voice assistant.

Top 5 projects from the buildathon

Nox Agent

Source code - https://github.com/satya-sudo/NoxAgent

Satyam Shree built a local-first, voice-controlled assistant for tasks such as music, search, reminders, and smart-home control. Core speech and conversational processing can run on user-controlled hardware, giving the project a clear privacy and data-control angle.

Demo of Nox Agent - Local AI-powered voice assistant

Flinch

Source code - https://github.com/voxmastery/flinch

Built by Ganesh, Flinch gives AI agents a way to remember harmful outcomes and avoid repeating the same mistake. It applies a familiar lesson, learning not to touch a hot stove, to agents that can run commands or change code.

CRON-X

Source code - https://github.com/manasvig-cyber/CRON

Manasvi built an AI-powered security operations dashboard. It brings together logs, looks for suspicious behaviour and intrusions, and can trigger responses while keeping a human involved in critical decisions.

GTM Engineering Agent

Presentation - https://gamma.app/docs/GTM-Engineering-Agent--ubhun5w7ee3yb5o

Sujan and team built a data enrichment pipeline with multiple coding agents reviewing the work. Each reviewer brings a different perspective to the code and data, applying an AI-assisted review process to a Google business data workflow.

AI-Powered 24/7 Radio

Source code - https://github.com/Narendrareddygithub/ai-powered-24-by-7-radio

Narendra’s project is an autonomous audio newsroom. It monitors signals from the open internet, deduplicates and verifies developments, writes radio commentary, and prepares a YouTube broadcast. Platform onboarding rules held back the live broadcast during the hackathon, but the project still showed the ambition of an always-on newsroom.

Why Nox Agent won

The judges evaluated all five projects against three equally weighted criteria: usefulness and impact, completeness, and technical excellence. Nox Agent made a strong case across all three.

Its usefulness starts with a familiar need. People want the convenience of a voice assistant, but some also want more control over where speech and conversation data are processed. Nox brings that privacy-minded approach together with concrete tasks, including reminders, music, browser actions, and smart-home controls.

Completeness was another strength. Nox connected voice input to command handling, a dashboard, an API, scheduling, browser actions, and Home Assistant integration. This made it feel like an assistant people could try, rather than a single feature waiting for the rest of a product to be built.

The engineering supported that end-to-end experience. The Go project is organized into distinct components, with deterministic command handling taking priority and a local language model available as a fallback. Model-driven actions are constrained, and the project includes persistent scheduling, documentation, and tests across core modules.

Nox did not need to claim that every edge case was solved. Its roadmap also calls out areas for further work, including wake-word reliability and offline text-to-speech consistency. That clear product direction, paired with an implementation people could explore, made the submission especially compelling.

A strong finish, and a starting point

Five projects shipped in 12 hours, each with a different take on useful agents. Flinch explored how agents could learn from mistakes. CRON-X brought AI into security monitoring with a human in the loop. The GTM Engineering Agent applied multiple reviewers to data work. AI-Powered 24/7 Radio aimed to turn live signals into an audio newsroom. Nox Agent brought local, voice-driven assistance into reach.

Congratulations to Satyam Shree and Nox Agent, and respect to Ganesh, Manasvi and team, Sujan and team, and Narendra for shipping their projects. The buildathon is over; the repositories and demo are there to explore, improve, and build on.

Watched from the sidelines?
Join the next AI Agent Buildathon.