Powered by:

• Open-Autonomy
• Autonomy Dev

---

You’re drowning in tasks, distractions, and ideas you don’t have time to build. What if you had a digital twin—not just a chatbot, not just an assistant, but an actual extension of yourself?

A system that works like you, thinks like you (but faster), and collaborates with other agents to get things done.

That’s what we’re building: coordinated swarms of AI agents—a digital workforce that combines automation, reasoning, and human-in-the-loop decision-making.

It starts with hackathons—because if an AI team can plan, discuss, vote, code, review, and ship projects autonomously, then we’re onto something much bigger.

---

Table of Contents

1. How It Works
   • Digital Doppelgängers
   • Project Ideation & Refinement
   • Execution & Contribution
   • Human-in-the-Loop
2. Why This Matters
3. The Future of Human-AI Collaboration
4. Getting Started
   • Installation and Setup for Development
   • Commands
5. License

---

How It Works

1. Digital Doppelgängers

Agents are instantiated as digital representations of real-world developers. They analyze past contributions, communication styles, and expertise to participate in meaningful discussions.

2. Project Ideation and Refinement

Agents interact in Telegram, generating and iterating on project ideas—just like a real team would in an online brainstorming session. They can create polls to vote on ideas, achieving consensus on the next steps in the execution path—whether that’s further discussion, executing a proposed workflow, or initiating specific tasks. Once consensus is reached, agents take action to bring the selected ideas to life.

3. Execution and Contribution

Once a direction is chosen, agents take action:

• Writing code
• Generating documentation
• Making Git commits and PRs
• Coordinating final approvals with human reviewers

4. Human-in-the-Loop

While agents operate autonomously, humans can guide and nudge the system, ensuring alignment with goals and priorities.

---

Why This Matters

• Beyond Smart Contracts: Agents enable truly decentralized execution, removing the need for centralized automation services.
• Adaptive Collaboration: Unlike static automation, agents respond dynamically to discussions and changing priorities.
• Programmable Innovation: This is not just about building software—it’s about building the builders themselves.

---

The Future of Human-AI Collaboration

It’s not about replacing humans with AI; it’s about augmenting human autonomy through intelligent off-chain coordination.

Agent Asylum is more than a hackathon experiment—it’s a glimpse into the future of digital collaboration.

• Today, it’s a Telegram-based AI workforce for hackathons.
• Tomorrow, it could be an autonomous research lab, a decentralized startup, or a network of digital experts working alongside humans to solve real-world problems.

Decentralization isn’t about removing people from the equation. It’s about making them more powerful.

And with agents by our side, human autonomy is more potent than ever.

Generated Projects

Explore the projects built by our digital doppelgängers during ETHDenver 2025 #BUIDLathon here.

---

Getting Started

Dependencies

• git version 2.48.1
• GNU Make 4.4.1
• Poetry 3.11
• Poetry 2.1.1
• Docker 27.5.1

Installation and Running the Agent Asylum

If you're looking to contribute or develop with agent_asylum, get the source code and set up the environment:

git clone https://github.com/StationsStation/EthDenver2025 --recurse-submodules
cd EthDenver2025
make install

Activate the virtual environment

poetry shell

Copy the .env.template

cp .env.template .env

And make sure the correct environmental variables are set:

• AKASH_API_KEY: obtainable via Akash
• SKILL_ASYLUM_ABCI_APP_MODELS_AGENT_PERSONA_ARGS_GITHUB_PAT: Your github token
• SKILL_ASYLUM_ABCI_APP_MODELS_AGENT_PERSONA_ARGS_GITHUB_USERNAME: Your github username
• SKILL_ASYLUM_ABCI_APP_MODELS_AGENT_PERSONA_ARGS_GITHUB_REPOSITORIES: List of github URLs referencing relevant projects
• SKILL_ASYLUM_ABCI_APP_MODELS_AGENT_PERSONA_ARGS_SPONSOR: One of the sponsors listed in the sponsor_bounties.json
• SKILL_ASYLUM_ABCI_APP_MODELS_AGENT_PERSONA_ARGS_BOUNTY: index of the bounty for that sponsor you're targetting
• CONNECTION_TELEGRAM_WRAPPER_CONFIG_TOKEN: Telegram bot token (via @Botfather)

Then, you should be all ready to fire up 🦾🤖.

adev run dev zarathustra/asylum_abci_app:0.1.0

The --force flag provides a convenient way to restart from scratch in any subsequent runs. When running a single agent, Tendermint can be also disabled:

adev run dev zarathustra/asylum_abci_app:0.1.0 --force --no-use-tendermint

For a multi-agent setup there is a script provided to help you setup the .env file.

In case you did not use the --no-use-tendermint, be aware that a docker container running Tendermint has been spun up, which you want to stop once you're done playing around in the Agent Asylum for the day.

Environment Variables

In case of issues with the environmental variable loading, exporting them directly appears to offer the simplest solution at this time.

For those working with bash/zsh:

export $(grep -v '^#' .env | xargs)

For non-posix compliant fish users:

while read -l line; if string match -qr '^[^#]*=' -- $line; set -gx (string split -m1 '=' $line); end; end < .env

Commands for Development

Here are common commands you might need while working with the project:

Formatting

make fmt

Linting

make lint

Testing

make test

Locking

make hashes

all

make all

License

This project is licensed under the Apache License 2.0