For the complete documentation index, see llms.txt. This page is also available as Markdown.

Introduction

A framework for AI agents to run decentralized infrastructure locally

The Obol Stack is a local-first agent harness: a Kubernetes cluster on your laptop, a default AI agent (Hermes) with its own Ethereum wallet, dynamically-deployable blockchain networks, a Cloudflare tunnel for public exposure, and an x402 payment gateway so agents can charge for what they serve.

The thesis is simple: agents should be able to run real infrastructure, build something valuable on top of it, and sell access to it for micropayments — without asking permission and without standing up cloud accounts.

The Obol Stack is alpha software. For production validator deployments, use the Run a DV docs and dedicated infrastructure.

What's in the box

Obol Stack is a two-part system:

  1. obolup.sh — bootstrap installer that lays down pinned dependencies (kubectl, helm, k3d, helmfile, k9s) and the obol CLI.

  2. obol CLI — Go binary that drives everything: cluster lifecycle, the agent, networks, payment-gated services, and the tunnel.

The cluster runs entirely on your machine via k3d (Kubernetes in Docker).

Key features

  • Agent-firstobol stack up brings up a default Hermes agent with its own Ethereum signing wallet (backed by a remote-signer), a chat TUI, and a growing skill set. Talk to it with obol hermes chat.

  • Sell what your agent buildsobol sell demo deploys a payment-gated HTTP service in one command. Use it as the starting point for selling inference, indexed data, or any HTTP API for $OBOL or USDC micropayments.

  • Native $OBOL micropayments with sponsored gas on mainnet — when buyers pay in $OBOL on Ethereum mainnet, the Obol facilitator sponsors the on-chain settlement gas. Buyers sign an EIP-2612 permit off-chain and never need ETH. Sellers receive $OBOL directly to their agent wallet.

  • Multiple network support — sync local Ethereum nodes (mainnet, sepolia, hoodi), Aztec sequencers, and more. Built-in eRPC routes to public RPCs when no local node is present.

  • Public access — expose only the routes you choose (/services/<name>/* and discovery metadata) via a Cloudflare quick tunnel. Internal routes (frontend, eRPC) stay locked to obol.stack.

  • Unique deployments — every install gets a uniquely-namespaced deployment, so multiple stacks coexist on one machine.

CLI overview

Command
Description

obol stack init / up / down / purge

Cluster lifecycle

obol agent init / new / setup / sync / list / delete

Manage agent instances (default runtime: Hermes)

obol hermes chat / skills / config / ...

Pass-through to the in-cluster Hermes CLI

obol model setup / status

Configure LLM providers (Ollama, Anthropic, OpenAI, custom)

obol network list / install / sync / delete

Manage blockchain networks

obol sell demo / inference / http / list / status / register

Create payment-gated services and register on ERC-8004

obol app install / sync / list / delete

Install arbitrary Helm charts

obol tunnel status / setup / restart

Manage the Cloudflare tunnel (setup creates a permanent URL)

obol kubectl / helm / helmfile / k9s

Kubernetes tool passthroughs (auto-configured KUBECONFIG)

Default infrastructure

When you run obol stack up, the following services are deployed automatically:

Service
Namespace
Purpose

Hermes (default agent)

hermes-obol-agent

AI agent + dashboard, with its own Ethereum signing wallet

Traefik

traefik

Gateway API ingress controller

Cloudflared

traefik

Cloudflare tunnel connector for public routes

eRPC

erpc

Unified RPC load balancer (local nodes + public fallbacks)

Obol Frontend

obol-frontend

Web management dashboard (local-only)

Monitoring

monitoring

Prometheus + kube-prometheus-stack

LiteLLM

llm

OpenAI-compatible LLM gateway (Ollama, Anthropic, OpenAI, custom endpoints)

x402 verifier + ServiceOffer controller

x402

Payment gating + reconciliation of payment-gated services

Use it from Claude Code

The Obol team publishes a Claude Code plugin with skills for installing, operating, and selling on the Obol Stack.

Once installed, Claude Code can drive obol stack up, set up the agent, troubleshoot pods, and walk you through obol sell demo. Source: github.com/ObolNetwork/skills.

System requirements

Prerequisites

  • Docker 20.10.0 or later (daemon must be running)

  • macOS (Darwin) or Linux

  • amd64 or arm64 architecture

Resource recommendations

Component
Minimum
Recommended

CPU

4 cores

8 cores

RAM

8 GB

16 GB

Storage

50 GB

500+ GB (varies by network)

Architecture overview

Where next

  • Quickstart — install the stack, talk to your agent, and run obol sell demo.

  • Build a profitable Obol Stack — end-to-end: bounded archive node → index → paid service → specialized agent → listed on marketplaces.

  • Selling agent services — full orientation on the three sell shapes, x402 economics, ERC-8004 registration, and Telegram notifications.

  • Installing Networks — sync Ethereum / Aztec, including bounded archives via --since.

Need assistance?

If you have questions or encounter issues with the Obol Stack, head over to our Discord where a member of our team or the community will be happy to assist.

Last updated

Was this helpful?