ETHGlobal

ETHGlobal Brussels

Galadriel

$10,000

galadriel.com

galadriel_ai

About

Galadriel is the first L1 blockchain purpose-built for AI.

ETH brought devs smart contracts to build dApps. Similarly, Galadriel enables devs to build AI apps & agents like smart contracts — in Solidity, decentralized, and on-chain.

Galadriel is built on a parallel-execution EVM stack to bring state-of-the-art LLMs (Large Language Models) on-chain in a low-cost, low-latency, and verifiable manner through Galadriel’s teeML. This makes it possible for devs to experiment & build highly capable AI agents on-chain.

Prizes

🏆 Best on-chain AI agent(s) ⸺ $6,000

🥇

1st place

$3,000

🥈

2nd place

$2,000

🥉

3rd place

$1,000

Build and deploy on-chain AI agent(s) on Galadriel Devnet. To build AI agents on-chain, Galadriel makes it possible to call LLMs in Solidity smart contracts, however, an AI agent is more than just calling an LLM for a single response. All essential components of an AI agent are the following:

LLM: the core engine of any language-based AI agent [https://docs.galadriel.com/teeML-features]

system prompt: the initial text input to LLM that describes how the agent will operate, could also include a persona description for the agent, try Googling “agent system prompt”
[https://docs.galadriel.com/tutorials/agents#before-agents-single-llm-call]

memory: there are 2 types of memory - chat history & knowledge base (RAG) that give the LLM context on previous messages with the user and any domain-specific knowledge, that the base LLM has not been trained on [https://docs.galadriel.com/tutorials/rag]

tools: enable the LLM agent to interact with external environments, e.g. web search, code execution environment, image generation [https://docs.galadriel.com/tutorials/agents#adding-a-tool-web-search]

running in a loop: calling out the LLM concurrently inside a loop to feed in every previous output, enables it to reason about its actions or “thoughts” to complete a user-given task [https://docs.galadriel.com/tutorials/agents#example-run-with-an-llm-loop]

Galadriel makes these features available for devs to build AI agents on-chain.

Qualification Requirements

The project must be deployed on Galadriel Devnet.
Must include clear instructions on how to use it.
A short description is included that explains how Galadriel is used in the project.
The project must involve on-chain AI features by Galadriel. The team will manually check for this. Also, deploying just regular EVM contracts without any AI use case will not count.

When picking the winner Galadriel team considers 1) usability, 2) uniqueness, and 3) technical execution.

💎 Multi Agent System Bounty ⸺ $4,000

Build and deploy more than 3 on-chain AI agents, that are interconnected and can communicate with each other via a shared interface, on Galadriel Devnet.

Such systems are called multi agent systems, which could be implemented as Sims-like games where multiple AI agents roam in a shared environment, and interact as NPCs, or a multi agent system can be thought as a team of agents working together to solve a given task (think like a team in a company).

For this track, we'd expect most of the effort going into figuring out how the on-chain agents can be built to communicate with each other on-chain.

So the best is to start by 1st understanding how LLMs work on Galadriel, then go through a single on-chain Agent example (both available as tutorials in our docs), and after that think how to connect more than one agent for your specific use case.

Here's some extra materials about multi-agent examples from web2 world:

https://www.youtube.com/watch?v=pJwR5pv0_gs
https://github.com/py499372727/AgentSims
https://microsoft.github.io/autogen/docs/Use-Cases/agent_chat/

Qualification Requirements

The project must be deployed on Galadriel Devnet.
Must include clear instructions on how to use it.
A short description is included that explains how Galadriel is used in the project.
The project must involve on-chain AI features by Galadriel. The team will manually check for this. Also, deploying just regular EVM contracts without any AI use case will not count.

NB! When picking the winner Galadriel team considers 1) usability, 2) uniqueness, and 3) technical execution.