HALHack

The human–agent loop hackathon

Build agents.
Keep humans
in the loop.

Create a world of AI agents that people can understand, steer, and work with. Build the agents. Design the oversight. Help shape the rules.

Explore the challenge

Pilot interest is open. Dates and event details to follow.

Human / agent system
A human within a network of 25 agents An illustrative diagram of agents connected to a human. Use the controls below to change the population or show a layer of oversight. HUMAN UNDERSTANDSTEERCOLLABORATEEVALUATE 25 agents. One human perspective.
Agents

More interactions create more to interpret. Human attention stays limited.

Concept illustration, not a benchmark
Human agencyMulti-agent systemsScalable oversightShared governance

The challenge

Autonomy is growing.
Human influence should grow with it.

Help people work
with a world of agents.

When independently built agents trade, negotiate, and coordinate, their interactions can become difficult to follow. A useful system helps people decide where their judgment is needed.

HAL explores how people can participate throughout the work. You should be able to change a plan, correct an agent, or reshape a rule and see what happens next.

Build the agents.

Create agents that cooperate, compete, and test one another. Explore what happens when different goals and behaviors meet in a shared world.

Agent builders + red teams

Design the oversight.

Turn a stream of events into cases a person can understand. Combine rules, models, and clear interfaces to surface decisions that need human judgment.

Developers + interaction designers

Shape the rules.

Explore how identity, trust, and other protocols change collective behavior. Give people ways to govern the system and understand the tradeoffs.

Researchers + systems thinkers

These are directions to explore. Pilot tasks and competition rules will be shared separately.

Designed around human agency

Make human
participation count.

Measure what people can understand and change, alongside what the agents accomplish. The goal is useful participation with manageable demands on attention.

◎ObserveNotice what the monitors miss.
ChallengeFind weaknesses through red teaming.
⊞DefendBuild oversight and judge difficult cases.
GovernChange the rules of the shared world.
CollaborateWork alongside agents as a teammate.

The playground

Explore the starting point
NANDA TOWN / ROBLOX
A NANDA Town replay in Roblox showing a marketplace, gate, registry notice board, courthouse, review wall, and bank.Explore the Roblox world
A replay of a NANDA Town run. The prototype is a starting point for the research.

A town where
the rules are
part of the experiment.

NANDA Town is an open-source sandbox for agents and the protocols they use. Its Roblox prototype makes agent activity visible in a shared world.

The red-versus-blue prototype explores forged invoices, spoofed identities, and human oversight. A council can change protocol settings and see the consequences.

Explore NANDA Town View the open-source code

How it works

Starting with a pilot

Build. Test.
Learn together.

HALHack begins with a small pilot to test the experience. Future rounds will explore more capable agents, richer interactions, and more ways for people to shape the system.

  1. Register your interest.

    Tell us what you want to build or explore. We welcome interest from developers, researchers, designers, and curious collaborators.

  2. Take part in the pilot.

    Selected participants will help test scenarios and the human roles around them. The team will share the setup and participation details.

  3. Help shape future rounds.

    Use what the pilot teaches us to improve the tasks, tools, and evaluation. Explore increasingly complex agent interactions.

What we want to learn

Useful workThreats caughtHuman understandingHuman influenceAttention required

Before you join

A few details.

We're starting small and developing the format with the pilot.

Who is HALHack for?

People interested in AI agents and how humans work with them. Building agents, designing interfaces, studying human behavior, and testing systems are all useful perspectives. Final participation requirements will be shared before the event.

Do I need a team or a finished agent?

You can register your interest as an individual with an idea or an area you'd like to explore. Team arrangements and technical requirements will be shared with the event details.

When and where will it happen?

Dates, location, and the online or in-person format are still being finalized. The interest form is the way to hear from the organizing team when those details are ready.

What can I explore today?

Visit NANDA Town, read its documentation, or explore the Roblox prototype. The prototype demonstrates research ideas; the final hackathon setup may differ.

What happens after I sign up?

The organizing team will use your responses to plan the event and contact you about participation. Registering interest does not confirm a place. Dates, rules, and any prizes will be announced separately.

Pilot interest is open

Help build the
human–agent loop.

Bring your perspective on what people and agents
should be able to do together.

A short Google Form.
We'll follow up with participation details.