Collaborative Agent Hackathon - Stanford, CA, 2026

General
1

565cabd5-54ee-477b-970a-786b4468bf71
565cabd5-54ee-477b-970a-786b4468bf71800×800 162 KB

Collaborative Agent Hackathon - Stanford, CA, 2026

It is great to have you taking part in our Collaborative Agent Hackathon in Stanford! As participants, you will be among the first to get limited access to Flower Agent and explore what becomes possible when collaborative AI runs on SuperGrid.

This post is the main reference point for the day, with information about the schedule, tracks, technical resources, demos, and support.

You do not need to arrive with a team or a finished idea. We will help you form teams, choose a direction and turn your ideas into working projects during the day.

At a glance

Registration and a light breakfast start at 9:30 am. Lunch, refreshments and snacks will be provided. Project demos begin at 5:15 pm, and prizes are awarded at 6:45 pm.

Before you arrive

To help you get started quickly on the day, please complete the following:

  • Create or verify your Flower Account. If you do not already have one, please create your Flower Account. Please have the username registered to your account ready on the day.
  • Install a Python package and project manager. We recommend uv. If you already use another Python package or project manager, you can use that instead.
  • Read the Flower Agent documentation. You do not need to build a complete project in advance, but the tutorials will help you get started quickly. (Refer to the Warmup section below)
  • Join the Flower Slack workspace. Join Flower on the Slack workspace and then join the #hackathon_stanford_2026 channel. We will use this channel for announcements, support and questions.
  • Bring your laptop and cables. Bring your laptop, charger and any cables or adapters you need. Extension cords will be provided, but supplies may be limited, so please bring a suitable extension lead or power strip if you have one.

To use Flower Agent on SuperGrid, we will help with access and allow-listing at the start of the day. Your registered Flower Account username will help us get you set up quickly.

Warmup

Please spend some time familiarizing yourself with the Flower Agent workflow before you arrive. These resources will help you get started quickly:

You do not need to complete every tutorial. A quick look at the relevant sections will make it easier to choose a direction and start building on the day :rocket:.

Schedule

  • 9:30 - Opening, check-in and breakfast
    Meet other participants and share your interests and experience.

  • 10:00 - Welcome!
    Introduction to the event, the challenge, and the day’s practical arrangements.

  • 10:15 - Flower Agent technical demo
    A technical introduction covering collaborative agent architectures, safe coordination, self-improvement loops, and the available Flower resources.

  • 10:30 - Team formation, hacking starts!
    Meet collaborators and form teams of two to four, unless you have already formed a team. Define your project, design the agent interactions, and start implementing.

  • 12:30 - Lunch time

  • 13:00 - Build session
    Develop and test the system, refine the collaboration or self-improvement loop, and prepare the demonstration. Mentors will be available throughout.

  • 16:30 - Demo prep
    Submission reminder and time to start refining your final demos.

  • 17:15 - Demos begin
    Teams present their projects, followed by brief questions.

  • 18:00 - Dinner

  • 18:45 - Awards and closing remarks

  • 19:30 - Finish

The challenge

Flower Agent Harness

Showcase the collaborative aspect of Flower Agents running on SuperGrid. Your goal is to use the existing SuperGrid infrastructure and show how multiple Flower Agents can work together to solve problems, coordinate tasks, and achieve more than a single agent could on its own. You might build an agent chain in which multiple agents contribute to a result, or use multiple `AgentApp`s that share context and hand work between agents. Bonus points if you use our recently released Endeavor model.

Support during the day

Mentors will be available throughout the day to help with setup, ideas, and technical questions. Updates will be posted regularly in the #hackathon_stanford_2026 Slack channel. Throughout the day, you can ask questions in Slack or speak directly to any member of the Flower Labs team.

Flower AgentApp Templates

The Flower AgentApp and Flower Collaborative AgentApp are great starting points to start building your own Flower Agent. The Flower AgentApp receives the current prompt together with the conversation’s previous user and assistant messages. The Collaborative AgentApp comes with Grid tools enabled, allowing agents to sample other agents in a federation, send messages to them, and retrieve their responses. You can find more details on how to run the Flower Collaborative AgentApp across Flower SuperNodes, in this GitHub repo.

Configuring SuperNode’s model endpoint

For a SuperNode to access a model provider, you need to configure the following environment variables.

For models served via Flower AI:

export FLWR_MODEL_API_KEY="<YOUR-FLOWER-API-KEY>"

You can generate <YOUR-FLOWER-API-KEY> from directly from the flower.ai webpage by going through Profile -> Settings -> API Keys. For the name of the model you can follow the OpenRouter format, e.g., openai/gpt-5.6-sol.

For models served via the Nebius Token Factory:

export FLWR_MODEL_API_ENDPOINT="https://api.tokenfactory.tf-ca1.nebius.com/v1/responses"
export FLWR_MODEL_API_KEY="<NEBIUS-API-KEY>"

We will share the <NEBIUS-API-KEY> during the hackathon on the Slack channel.

Using Nebius Serverless AI and Token Factory Endpoints

Spinning up a Flower SuperNode on a Nebius Serverless AI endpoint

Log in to SuperGrid

Run flwr login supergrid

This will open a new browser tab and ask you to Grant Access to flower.ai Website.

Generate a SuperNode key pair

Run this command on your local machine:

ssh-keygen -t ecdsa -b 384 -N "" -f nebius-supernode-key-1

which will create: nebius-supernode-key-1 and nebius-supernode-key-1.pub

Register the public key with SuperGrid

Run: flwr supernode register nebius-supernode-key-1.pub supergrid

Create a custom endpoint in Nebius

From the Nebius console, navigate to:

  • Serverless AI → Endpoints → Create Endpoint

Then use the following values to populate the form

Field Value
Name <YOUR_ENDPOINT_NAME>
Configuration Custom
Image path docker.io/flwr/supernode:1.37.0-py3.12-ubuntu24.04
Private registry Leave empty
Port 9092
Protocol TCP
Entrypoint command exec flower-supernode --superlink=fleet-supergrid.flower.ai:443 --auth-supernode-private-key=/tmp/nebius-supernode-key-1 --allow-runtime-dependency-installation
Entrypoint note Make sure the Entrypoint command is entered as a single line.
Environment variables Leave empty
Secret environment variables Leave empty
Bearer token authentication Off
Platform - Instance With GPUs, NVIDIA L40S if the ClientApp trains or runs inference on GPU
Platform - VM type Regular
Settings - Preset CPU/RAM sized for your ClientApp. The default 1 GPU - 8 CPUs - 32 GiB RAM should be sufficient.
Storage - Container disk size 250 GB
Storage - Mounted volumes Leave empty
Files Upload the private nebius-supernode-key-1 key and set the mount path to /tmp/nebius-supernode-key-1
Access - SSH key Optional, but useful if you want to connect to the instance
Network - Endpoint type Select Public, or use a Private subnet with outbound internet/NAT access

Confirm the SuperNode is online

From your local machine, run:

flwr supernode list supergrid --verbose

The SuperNode should now appear as online.

Add more SuperNodes

To register additional SuperNodes, repeat the above steps for each new SuperNode.

Nebius Token Factory Available Endpoints
Model FLWR_MODEL_API_ENDPOINT Model ID API Key
Kimi-K2.7-Code https://api.tokenfactory.tf-ca1.nebius.com/v1/responses dedicated/flowerai/Kimi-K2.7-Code-1OUHWL Shared in Slack
MiniMax-M3 https://api.tokenfactory.tf-ca1.nebius.com/v1/responses dedicated/flowerai/MiniMax-M3-OOLI9o Shared in Slack

Submission and demos

Before the demos, each team will submit:

  • Team details: team name, members, and email addresses. You can register using this form.
  • Published Flower Hub app. You can publish your AgentApp following these instructions.
  • Short project description
  • GitHub repository link

Each team will have 3–5 minutes to present what they built, followed by questions from the judges. Keep your demo focused on:

  • How Flower Agent and SuperGrid enable your solution
  • The problem you are solving
  • The working result

Prize evaluation criteria

Agent performance may inform the assessment, but it is not the sole or decisive factor. Projects will be assessed around the primary theme of Collaborative Agents, based on the following dimensions:

  • Use of Flower - How much does the project uses Flower Agents and SuperGrid.
  • Impact & Originality - What is the unique value the project creates.
  • Demo & Delivery - How clearly the project’s approach, and results are presented.

Partners and prizes

We are delighted to be partnering with Nebius, ARM and AMD.

Prizes will be awarded to the top three projects at the end of the day, with a total prize pool worth up to $5,000.

We are excited to see what you build! Come ready to experiment, learn from one another, and explore what collaborative agents can do with Flower.

See you in Stanford!

The Flower team

FAQ

How many people should be on a team?

We recommend teams of 3–5 people, ideally with complementary interests and experience.

Can I ask questions to the Flower team?

Yes! We are more than happy to answer any questions you may have. Just walk up to any Flower Labs team member and ask directly.

Is there a credit limit for running agents on SuperGrid?

Yes. We will ensure that you have sufficient credits to run your AgentApps on SuperGrid throughout the duration of the hackathon.

Speak to any member of the Flower Labs team if you think you will need additional credits to complete your project during the event.

Is there a time limit for running AgentApps on SuperGrid?

Yes. Each task has a 5-minute timeout, starting from when the task switches to the Running status.

I have some feedback! Where/who can I share it with?

That’s awesome, and thank you for that! We welcome any feedback to improve the overall Flower Agent experience.

You can share your feedback by clicking the “Share feedback” icon on SuperGrid. Alternatively, you can share it in the #hackathon_stanford_2026 channel or speak directly with one of the Flower Labs team.

565cabd5-54ee-477b-970a-786b4468bf71
565cabd5-54ee-477b-970a-786b4468bf71800×800 56.1 KB

2 Likes