Building in Denver: Gaia enables developers to run open-source LLMs on their own hardware

Harish Kotra, a leading voice on Gaia’s developer relations team, recently took the stage at the Eigen Games Hacker House during ETHDenver 2025. 

Couldn’t make it to the IRL event? Here are the main takeaways and practical insights from his keynote:

Gaia Nodes: What You Get Out of the Box

We created Gaia to give developers the ultimate toolkit for building in decentralized AI. By allowing devs to leverage any framework, knowledge base, or open-source LLM, Gaia provides a set of flexible rails to locally deploy custom AI applications in just a few lines of code. 

• Wallet Integration — Every Gaia node is deployed with its own wallet address, making it possible for AI agents to autonomously perform on-chain transactions.
• Tool Usage & Function Calling — Because large language models (LLMs) rely on external data for more advanced tasks, Gaia supports function calling to third-party APIs — be it blockchain explorers, price feeds, or custom REST endpoints.
• Flexible LLM Choices — Although nodes ship with Llama 3.2 by default, you’re free to switch to other open-source models such as Qwen or Mistral.
• Embedded Vector Database — By default, Gaia uses Qdrant for vector indexing. If you prefer Postgres with PG Vector, the node can be reconfigured to suit your storage or retrieval needs.
• WasmEdge Runtime — Gaia’s reliance on WasmEdge ensures broad compatibility, whether you’re spinning up a node on a Raspberry Pi, personal laptop, or a cloud VM.

OpenAI-Compatible API

A highlight of Harish’s talk was Gaia’s OpenAI-compatible endpoint, allowing developers to plug in any ChatGPT or embeddings-based workflow with a simple URL switch. 

This compatibility makes it straightforward to integrate Gaia with existing tools or frameworks, reducing friction for those already familiar with OpenAI’s services.

Potential Projects & Use Cases

1. Digital Twins — Create AI agents that mimic well-known figures or domain experts—Harish mentioned existing demos of “Vitalik” and “Diplo” Digital Twins.

2. Localized Code Models — Fine-tune your Gaia node for generating and reviewing code in languages like Rust or Python, then incorporate it into your IDE.

3. Blockchain Interactions — Leverage the built-in wallet for fully autonomous on-chain deployments. Harish showcased examples of Gaia nodes launching tokens on chains like Celo or Base.

4. GitHub Repo Chat — Dig into codebases by dropping them into a Gaia-backed interface. One demonstration focused on letting an AI “judge” hackathon projects by analyzing the participants’ repos.

Code Samples & Demos

Throughout the keynote, Harish referenced a range of GitHub repos:

• Local PDF Q&A — A quick setup for dropping PDFs into a locally running Gaia node for question-answering.
• Node Launching via NPM — Spin up multiple Gaia nodes programmatically—each hosting a different open-source LLM.
• Tool-Calling Examples — Interacting with third-party APIs (e.g., Zerion) to fetch historical blockchain data.
• Telegram Bot — Hook your Gaia node to Telegram for real-time AI chat.

If you’d rather not run your own infrastructure, Harish pointed out Gaia’s public nodes running large models like Llama 70B or Qwen 72B that are ready to use.

Next Steps for Builders

• Visit Gaia’s Documentation — Check out docs.gaianet.ai to learn how to initialize a node, configure different models, or connect external data sources.
• Try Out the GitHub Projects — From PDF chat to on-chain token deployments, Harish’s repos provide hands-on templates for building AI-driven apps on Gaia.

Harish concluded by noting that Gaia serves as the “LLM inference layer,” ready to integrate with any front-end or workflow tool. 

Whether you’re launching Digital Twins, diving into code analysis, or interacting with on-chain assets, Gaia’s decentralized architecture ensures both flexibility and control — putting you in charge of how and where deploy your AI.

Want to hear more from Harish? Check out his presentation at Hyperbolic's Hype(r)House ETHDenver here: