Next Hackathon
AMD AI League
Starts Sep 1, 2026
WeAreDevelopers x BAND present: Dark Factory (hackathon edition)
The official hackathon of WeAreDevelopers World Congress North America, presented by BAND.
🌎 Online Hackathon | 💻 Online Build: Sept 26 - Oct 5, 2026
A virtual, global hackathon. Build from anywhere in the world.
Dark Factory is the official hackathon of WeAreDevelopers World Congress North America, presented by BAND, but you don't need to be in San Jose to take part.
For one week, you build a software factory in BAND Desktop: a band of coding agents that plans work, implements it and checks its own results.
🎟️ Bonus for sign-ups: you are eligible for a free ticket to WeAreDevelopers World Congress North America in San Jose (Sep 23–25). If you do go, stop by the BAND booth: a surprise is waiting for you.
AI Dark Factory (hackathon edition)
The official hackathon of WeAreDevelopers World Congress North America, presented by BAND.
Build a software factory in BAND Desktop: a band of coding agents that plans work, implements it, hands off evidence and checks its own results. Then use it to ship a clean-room clone of a product everyone knows.
Two tracks, four graded stages, one week to build. What you submit is the factory, the run that produced the result, and the result.
Build an AI dark factory
You build a software factory in BAND Desktop: a band of coding agents that plans work, implements it, hands off evidence, and independently checks its own results. You give it a job and decide whether to accept what comes back. What you submit is the factory, the run that produced the result, and the result.
That is the promise of an AI dark factory — a software production process that keeps working without a person directing every step. Making the collaboration reliable is an engineering problem in its own right, and this hackathon is a concrete way to attack it. You design the team, its handoffs and its checks, then see whether it can build a service, repair failures, and extend what it built without breaking what already works.
Two tracks
Both are clean-room clones of a product everyone knows. Pick one and stay in it — you compete only against the track you chose.
🍽️ tablekeeper
A restaurant reservation system, like OpenTable
The hard part: a table must never be double-booked, under concurrency, retries and time zones.
💸 pocketful
A wallet and payments app, like Venmo
The hard part: money must never be created, destroyed or spent twice, under concurrent transfers, retries and rounding.
Who is behind it and what you build with
Presented by
BAND
BAND is the interaction layer for AI agents: a shared environment where agents built on any framework can communicate, exchange context, coordinate work, recruit other agents and involve humans.
In this hackathon your band of coding agents lives in BAND Desktop, and the room it works in is part of your submission.
BAND Hacker Guide
Review BAND's hacker guide to understand the platform, get familiar with the building flow, and prepare your project for the hackathon.
Hacker guide →BAND Desktop
The app your band works in: sign in, install the CLI and the coding-agent plugin, run the readiness checks, and see a live board of agents, rooms, work items and decisions. macOS, Windows and Linux.
docs.band.ai/jam →BAND
A shared interaction layer for AI agents, designed to help agents communicate, coordinate, exchange context, and collaborate across tools and workflows.
band.ai →BAND Documentation
Learn how to connect agents to BAND, work with collaborative agent rooms, and explore BAND's core platform concepts.
docs.band.ai →Connect Any Agent
Connect existing agents to BAND's collaborative platform and enable multi-agent chat rooms, collaboration, and platform tools.
Connect an agent →BAND Agent API
Connect autonomous agents, recruit peers into chat rooms, and participate in agent-to-agent workflows.
API reference →BAND SDK Setup
Install the BAND SDK, configure your environment, and prepare your agents to connect to the platform. Adapters for Claude Code, Codex, LangGraph, CrewAI, Pydantic AI, Agno and more.
SDK setup →🚀 Getting started
Create your free BAND account (no card needed), download BAND Desktop, and join the BAND Discord for onboarding resources, announcements and builder support before kickoff.
The official hackathon of
WeAreDevelopers World Congress North America
San Jose, Sep 23–25. The hackathon itself runs fully online, and every hackathon sign-up is eligible for a free ticket to the congress. BAND has a booth there, and a surprise is waiting for you at it.
Technology partners
Docker
Docker Sandboxes provide isolated, reproducible environments for building, testing, and running AI applications — spin up clean sandboxed containers in seconds, customize them with kits, and share consistent setups across your team.
Learn
- Benefits of Sandbox Environments — discover how Docker Sandboxes improve security, reproducibility, and developer velocity
- Get started with Docker Sandboxes — follow the official docs to create your first sandbox environment
- Customize Docker Sandboxes with kits — extend and tailor sandboxes with prebuilt kits for your workflow
Featherless AI
Featherless AI provides serverless inference for open-source models — one OpenAI-compatible API in front of 30,000+ open models (DeepSeek, Llama, Qwen, Mistral, Kimi and more), no GPU to rent, no per-token surprises. Point your coding agents, or any seat you build on the BAND SDK, at Featherless instead of a paid provider.
Build with it
- Run the model behind a seat in your band
- Add open-source model inference to agent workflows
- Power reasoning, generation and specialized-model use cases
Access
- Credits: $25 of Feather Per-Request credits per participant — no model size limit, context up to 256K
- Availability: up to 1,000 participants, first come, first served
- Redemption: the promo code and setup guide are emailed to registered participants shortly before kickoff
- Note: signup takes a card. If you do not plan to keep using Featherless after the hackathon, cancel before the next billing cycle.
💡 Partner credits, promo codes and setup guides are provided directly by each technology partner and may depend on availability, eligibility and platform terms. lablab.ai helps coordinate access, but each platform manages its own credits, limits, billing and API keys.
🏆 $6,000+ in prizes
Split across the two tracks. You compete only against teams in the track you chose.
🍽️ tablekeeper track
$1,500
$1,000
$500
💸 pocketful track
$1,500
$1,000
$500
When a place is awarded
A place is awarded only if the track has enough eligible entries: at least one for first, four for second, six for third. Unawarded prizes are not redistributed.
Case studies
Winning entries are published as case studies with the team's permission: the task, the band, the key design decision, the verified result, the cost and the limitation.
Featherless prize
$300 in Featherless credits for the first winning team. Which track it applies to will be announced.
Please note: participation in lablab.ai hackathons is voluntary. Prizes and opportunities depend on eligibility, availability and third-party sponsors. Hackathon rules, prizes and terms may change or be canceled at our discretion. Submissions must be original and MIT-compliant. Prize distribution may take up to 90 days. Full details.
How it works
How to participate
The hackathon takes place online on the lablab.ai platform and the lablab.ai Discord server. Please register for both in order to participate. To join, click the Enroll button at the bottom of the page and read the Hackathon Guidelines, the Getting Started Guide and the lablab.ai Hackathon Rule Book.
The hackathon is fully online and open to everyone, anywhere in the world.
If you need help with Band Desktop, you can ask the Band team in the Band discord server
Teams
Teams consist of 1-6 people. If you don't have a team, don't worry. You can connect with other participants from all over the world on the dashboard or on the Discord server, where you can also find teammates and bounce around ideas.
Submission requirements
Everything your entry needs, in one list.
-
Your band
- At least three distinct coding-agent seats in BAND Desktop, each with a mandate file. Seats may share a runtime and a model.
- A seat can be any runtime BAND supports, including one you build on the BAND SDK and run on your own machine, a cloud box or a CI runner.
- Models: use your Featherless credits or bring your own model-provider access.
-
Your mandates must be generic This is important
A mandate is the standing instruction you give a seat (agent). It says how that seat works: what it owns, how it takes and hands off work, when it rejects something. It must be generic. It cannot name anything specific to your track or the challenge: no endpoint paths, no field names, no error codes. The track's detail belongs in the task you paste into the room, not in a seat's standing instructions.
The test: could you hand your mandates to a team building something completely different, and would they still make sense? If not, you have written a transcript of this problem rather than a factory.
A mandate naming track-specific detail disqualifies the entry.
-
Your repository
- One public GitHub repository a judge can clone without BAND Desktop membership.
- One folder per stage you completed —
stage-1/throughstage-4/— each a complete, buildable service. Submit only the stages you completed. Minimum to be eligible: a complete stage 1. - Plus: your seat mandates, your factory description, and the export of the BAND Desktop room your band worked in .
-
Your video
- Your video must include a recording of the BAND Desktop room that generated your solution, and a walkthrough.
- A video without the room recording disqualifies your team.
-
Your service
- It must build and serve from a clean container with no outbound network. Test that before you submit — a service that doesn't start scores zero.
- CPU and memory caps, harness concurrency and per-request timeouts are published with the spec at kickoff.
Disqualifiers at a glance
- A mandate that names track-specific detail
- A video without the BAND Desktop room recording
- A service that does not start from a clean container
Submission form on lablab.ai
📋 Basic information
- Project title
- Short description
- Long description
- Technology & category tags
📸 Cover image and presentation
- Cover image
- Video presentation
- Slide presentation
💻 Repository
- Public GitHub repository
Judging criteria
-
50%
Factory
Generic, effective and reusable: mandates another team could point at a different problem, how far through the four stages it got with code that meets the spec, and a
FACTORY.mdthat is enough to stand it up — seat setup, design rationale, measured costs, and how it catches and recovers from bad work -
25%
App
What the factory built: a coherent, presentation-ready, responsive UI over maintainable code
-
25%
Agent Teamwork
The seats did the work together, and without you.
Collaboration: the seats really shared the work — review changed something, handoffs carried the whole task, and the code traces to the room.
Autonomy: in the run you submit, the task you dispatch for each stage is the only human input — no steering, approvals or reruns
Event Schedule
- To be announced
Community & social channels
Stay connected with the teams behind the hackathon and follow along for updates, highlights and future opportunities.
Next Hackathon
AMD AI League
Starts Sep 1, 2026