What would you build if someone handed you $250 of cloud compute for AI agents?
That's not a hypothetical. Docker is giving every new Docker Agentic Platform signup a $250 Cloud Sandbox credit. That's enough to run 4 agents, 7 hours a day, for a whole month.
So we're turning it into a hackathon. 🐳
The $250 Agent Challenge is an open, online hackathon run by the Collabnix community. Claim the credit, build something real with it, and show us what an agent can do while you sleep. Winners get announced on stage at our Meetup in Bangalore on 21 November.
Why this is worth your weekend
AI agents are moving off our laptops. AWS, Microsoft and Docker now all run agents the same way:
- 🧱 every agent gets its own microVM
- 🔑 the agent never holds your real API keys
- 🔌 tools come in through a gateway (MCP)
- 🧯 outbound network is locked down by default
- ⏰ agents run on a timer while nobody is watching
Docker Cloud Sandboxes give you all five, with the same sbx CLI on your laptop and in the cloud. This hackathon is your chance to try that pattern hands-on, without an enterprise contract.
Pick a track
| Track | Build an agent that... |
|---|---|
| 🌙 Night Shift | does real work while your laptop is closed |
| 🔐 No Keys Allowed | uses real APIs but never sees your tokens |
| 🔌 Tool Master | connects MCP servers to do something useful |
| 👥 The Crew | runs several agents in parallel, each in its own sandbox |
| 🧰 Kit Maker | ships as a kit others can run with one command |
Solo or teams of up to 3.
Special awards: 💰 Best Bang for the Buck (best build for the least credit spent) · ✍️ Best Build Log · 🙌 People's Choice
Get started in four commands
# 1. Install (Windows: winget install -h Docker.sbx)
brew install docker/tap/sbx
# 2. Sign in
sbx login
# 3. Store your model key for the cloud (the agent only ever sees a placeholder)
sbx --cloud secret set anthropic
# 4. Run an agent in the cloud
sbx --cloud run claude --name my-hack
That's the whole loop. Claude Code, Codex, Copilot and Gemini all work out of the box.
Tip: set an auto-stop timer on every sandbox so nothing runs up your credit while you sleep.
No credit card? You can still compete. The same sbx CLI runs locally, and local entries are fully welcome. 💰 Best Bang for the Buck is the only award judged on cloud spend.
What you submit
Four things, by the deadline:
- A build log (a blog post or a repo README) that lets anyone repeat your build.
- A 2-minute demo video. Record it and share the link; no live demo needed.
- A screenshot of your spend, with private details blanked out.
- The submission form.
Then the community votes on the recorded demos.
The timeline
| When | What |
|---|---|
| Sat, Oct 10 | Online kickoff: walkthrough and Q&A |
| Oct 10 to Oct 31 | Build |
| Sat, Oct 31, 11:59 PM IST | Submissions close |
| Nov 1 to Nov 14 | Judging and People's Choice voting on the recorded demos |
| Sat, Nov 21 | 🎉 Winners announced at our Meetup in Bangalore, with Docker swag and goodies |
⏰ Claim the credit early. Docker's offer ends Oct 31, 2026, 11:59 PM PT. It's one credit per account, and a credit card is needed at signup.
Everything lives in the repo
Rules, FAQ, the getting-started guide and the build log template:
ajeetraina
/
250-dollar-agent-challenge
💰 The $250 Agent Challenge — an open hackathon: claim $250 in Docker Cloud Sandbox credit and build an AI agent that works while you sleep.
💰 The $250 Agent Challenge
An open hackathon where you claim $250 in Docker Cloud Sandbox credit, build something real with it, and show the community what an AI agent can do while you sleep.
The deal
Docker is giving new Docker Agentic Platform signups a one-time $250 compute credit (10× the standard $25) for its Cloud Sandboxes: microVMs that each run their own kernel in the cloud, so your agents keep working after you close your laptop.
What $250 actually buys you (Docker's own example, priced at the Medium tier of 4 vCPU, 8 GB, $0.28/hr):
- 🧑🤝🧑 4 agents in parallel, 7 hours a day, for 30 days, or
- ⚡ ~3,500 fifteen-minute automated runs
⏰ The offer runs through October 31, 2026, 11:59 PM PT. A credit card is required at signup, and it's one credit per user / account.
👉 Claim it: …
Ready?
- ⭐ Star and read the repo: github.com/ajeetraina/250-dollar-agent-challenge
- 💳 Claim your $250: docker.com/c/sbx-promo
- 📝 Submit when you're done: the submission form
Come build in the open with us. Show the community what a month of agents can really do, and we'll see the winners on stage in Bangalore on 21 November.
Drop your idea in the comments 👇 Teams often form right here.
Top comments (1)
Việc chạy AI agent trong các sandbox cô lập như Docker Cloud thực sự là một giải pháp hay để kiểm soát rủi ro khi agent có khả năng thực thi code. Một vấn đề mình thường gặp là quản lý tài nguyên và giới hạn API quota khi chạy nhiều agent cùng lúc trong môi trường sandbox. Nếu không tính toán kỹ, chi phí cloud sẽ đội lên rất nhanh vì các tiến trình chạy ngầm hoặc vòng lặp vô tận của agent. Thay vì đổ hết tiền vào compute, mình thấy việc tối ưu hóa cách agent gọi model qua API quan trọng hơn để giữ cho ngân sách $250 đó thực sự hiệu quả cho việc scale sản phẩm — found it via LabAgent, site: labagent .tech