Release v0.7.0-rc0

v0.7.0-rc0 — Agent Commerce

This release candidate turns the Obol Stack into a self-monetizing agent platform: register on-chain, sell inference for USDC, discover other agents, and buy their services — all from a single obol stack up.

Highlights

• Zero-config setup — obol stack up is now a single command that produces a fully working, publicly accessible, payment-gated inference stack
• Agent Registry dashboard — browse on-chain registered agents from the ERC-8004 Identity Registry, filter by x402 payment support or protocol (MCP, A2A, Web)
• Per-MTok pricing — sell inference with token-based pricing (--per-mtok), not just flat per-request
• Wallet backup & restore — obol openclaw wallet backup / restore for remote-signer key management
• 60-70% token savings — optimized heartbeat reduces daily LLM consumption through compact output and throttled discovery

What's New

Zero-Config Stack (obol stack up)

No more manual steps. A single command now:

1. Deploys infrastructure (Traefik, LiteLLM, x402-verifier, CRDs)
2. Auto-configures LiteLLM with local Ollama models
3. Sets up OpenClaw agent with qwen3.5:9b as default model
4. Starts a Cloudflare tunnel with URL injection into the agent

All OpenClaw inference routes through LiteLLM — no more "No API key for anthropic" errors when importing existing configs.

Agent Registry (Frontend v0.1.12)

New collapsible Agent Registry section on the dashboard, powered by the 8004scan public API:

• Rich agent cards with avatar, name, description, owner, service endpoints
• x402 badge — identifies payment-gated inference services at a glance
• Filter chips — x402 (client-side), MCP, A2A, Web (server-side via 8004scan)
• Free-text search + auto-refresh every 30s
• Sell Inference modal — create ServiceOffer CRs from the browser (no CLI needed)

Payment-Gated Inference

Full commerce loop validated end-to-end: sell → 402 → EIP-712 sign → x402-rs facilitator → USDC settled → 200 + inference.

• obol sell http with --per-mtok and --per-request pricing
• Static paid/* LiteLLM route through x402-buyer sidecar
• Pre-signed ERC-3009 authorization pool (bounded spending, zero signer access)
• Works both via obol.stack:8080 and through Cloudflare tunnel

ERC-8004 Agent Identity

• On-chain registration on Base Sepolia Identity Registry
• OASF-enriched .well-known/agent-registration.json with skills/domains taxonomy
• On-chain metadata: x402.supported, service.type
• discovery.py search --x402-only for indexed agent discovery
• CLI flags: --register-skills, --register-domains

Heartbeat Efficiency (#263)

~60-70% reduction in daily LLM token consumption:

• monetize.py process --all --quick — single-line compact output
• Buy-side discovery throttled to ~every 10 minutes (was every heartbeat)
• buy.py maintain replaces separate balance + list calls
• Heartbeat interval 1m → 2m

Wallet Management (#260)

• obol openclaw wallet backup <id> — encrypted backup with AES-256-GCM
• obol openclaw wallet restore <id> --input <file> — restore + restart remote-signer
• Passphrase-protected or plaintext backup modes

Infrastructure

Component	Version
OpenClaw	v2026.3.11
Frontend	v0.1.12
obolup.sh	OpenClaw 2026.3.11

BDD Integration Tests

7 Gherkin scenarios, 75 steps — real user journey with no kubectl shortcuts:

1. Operator sells inference via CLI + agent reconciles
2. Unpaid request → 402 with pricing
3. Paid request → 200 + real inference
4. Discovery-to-payment cycle (402 → parse → sign → pay)
5. Paid request through Cloudflare tunnel
6. Agent discovers service through tunnel
7. Operator deletes ServiceOffer + cleanup

Upgrade

OBOL_RELEASE=v0.7.0-rc0 ./obolup.sh
obol stack init && obol stack up

What's Changed

• feat: BDD integration tests + upstream auth injection + tunnel URL by @bussyjd in #256
• feat: zero-config stack setup + OASF-enriched ERC-8004 registration by @bussyjd in #257
• Reflect user feedback by @OisinKyne in #259
• Adds obol openclaw wallet backup and obol openclaw wallet restore by @bussyjd in #260
• Add static paid x402 gateway and per-MTok pricing support by @bussyjd in #262
• perf: optimize heartbeat for token efficiency by @bussyjd in #263
• Bump OpenClaw to v2026.3.11
• Bump frontend to v0.1.12 (Agent Registry + Sell Modal)

Full Changelog: v0.6.0...v0.7.0-rc0

v0.6.0 Local First Agents

v0.6.0 — Local Agents for Ethereum

This is the Obol Stack's v0.6.0 release. It allows you to run AI agents on your own hardware with local-first inference and decentralised infrastructure.

Highlights

• Qwen 3.5 as default local model — obol model setup now also recommends qwen3.5:9b, a strong general-purpose model that runs well on consumer hardware. (~12GB of RAM, up to 256k context)
• LiteLLM gateway — replaced llmspy with LiteLLM, an industry-standard OpenAI-compatible proxy with wildcard routing, provider fallbacks, and broad model support
• Smart model hierarchy — cloud models (Anthropic, OpenAI) automatically become the primary when added via obol model setup, with local models as fallbacks
• Sell-side monetization — obol sell enables pay-per-request access to local compute via x402 micropayments (USDC on Base Sepolia)
• Buy-side payments — x402-buyer sidecar with pre-signed ERC-3009 authorizations for bounded, zero-signer-access spending
• On-chain agent registration — ERC-8004 Identity Registry on Base Sepolia with obol sell register
• Agent discovery skill — agents can search the ERC-8004 registry to find and purchase services from other agents
• Base Network Support - Adds Coinbase's Base network as a default free RPC

What's New

LLM Routing

• LiteLLM with wildcard routing (anthropic/*, openai/*) — add a provider once, use any model
• obol model sync pushes LiteLLM model list to all OpenClaw instances via host-side file patching (hot-reload, no pod restart)
• obol model setup auto-detects Ollama models and registers them in LiteLLM
• Tool calling support through LiteLLM for Ollama models

Monetize (x402)

• Full sell → discover → buy → settle roundtrip tested end-to-end
• ServiceOffer CRD with 6-stage reconciliation (monetize.py)
• x402-verifier ForwardAuth middleware for Traefik Gateway API
• Buyer sidecar with pre-signed authorization pool

Infrastructure

• Pinned Ethereum client versions for reproducible deployments
• Local networks registered in eRPC with proper aliases
• Disabled default public tunnel (opt-in via obol tunnel provision)
• Remote-signer chart bumped to 0.3.0
• OpenClaw chart 0.1.7 with LiteLLM defaults and model fallback support

CLI

• obol sell — inference, http, list, status, stop, delete, pricing, register
• obol model sync — refresh OpenClaw model lists from LiteLLM
• obol model pull — interactive model selection with pull progress
• Renamed: obol rpc → obol network, obol monetize → obol sell

Upgrade

OBOL_RELEASE=v0.6.0 ./obolup.sh
obol stack init && obol stack up

# To upgrade from a local model to a remote anthropic or openAI model
obol model setup

What's Changed

• Local networks in erpc with proper alias by @OisinKyne in #247
• Fix ethereum chart, pin version by @OisinKyne in #248
• Pin Ethereum client versions by @OisinKyne in #249
• feat: replace llmspy with LiteLLM as LLM gateway by @bussyjd in #243
• Update llm selection to ensure if you add a remote one it takes priority by @OisinKyne in #252
• Upgrade just one chart by @OisinKyne in #242
• Fix flicker and password prompt by @OisinKyne in #253
• Add Obol's Base env by @OisinKyne in #254
• Fix unbound variable crash by @OisinKyne in #255

Full Changelog: v0.5.0...v0.6.0

Release v0.5.0

v0.5.0 — Selling on Ethereum

This is the Obol Stack's v0.5.0 release. This release begins the support for x402 payments on Base Sepolia testnet. An upcoming v0.6.0 will roll out wider support on mainnet and Base mainnet, and a further update will add flexibility to the assets that you can pay in, and resources you can pay for.

Get started with the quickstart guide here.

Warning

The Obol Stack is in early alpha, please expect breaking changes. Do not use the stack with must-work applications, nor should you give Obol Agents more crypto that you are comfortable with them losing. The x402 feature is very early and under-documented, YMMV.

Upgrading

bash <(curl -s https://stack.obol.org)
obol stack up
obol openclaw sync

What's Changed

• Fix OpenAI inference through llmspy by @bussyjd in #214
• feat: secure enclave inference, x402 monetization, and ERC-8004 registration by @OisinKyne in #218
• Bump remote signer chart by @OisinKyne in #240
• Simplify erpc by @OisinKyne in #241
• Prep for a change to upstream erpc by @OisinKyne in #233
• Updates for rc3 by @OisinKyne in #244
• Disable default public tunnel by @OisinKyne in #245
• Fix network installation by @OisinKyne in #246

Full Changelog: v0.4.0...v0.5.0

v0.4.0 - Writing to Ethereum

v0.4.0 — Writing to Ethereum

Every Obol OpenClaw agent now ships with its own Ethereum signing wallet. During obol openclaw onboard (or obol agent init), a secp256k1 key is generated, encrypted as a V3 keystore, and provisioned to a per-instance remote-signer service running in the same namespace.

Agents can sign transactions, messages (EIP-191), typed data (EIP-712), and raw hashes via the ethereum-local-wallet skill — without ever touching private key material directly. Future updates will introduce multi-signature and smart-wallet functionalities, to further improve your Agents ability to serve you securely on chain.

Get started with the quickstart guide here.

Warning

This wallet feature is in early Alpha, please do not use more than a small amount of gas money with this wallet. Any significant mainnet funds should be at least joint-custodied with a human in the near term.

Foundry Tools Baked into OpenClaw Image

cast, forge, and anvil are now pre-installed in the OpenClaw Docker image, replacing the previous init container approach that copied binaries at pod startup. This eliminates startup latency and PATH override hacks.

• New docker/openclaw/Dockerfile layers Foundry binaries onto the upstream OpenClaw base image
• The CI workflow (docker-publish-openclaw.yml) now builds in two stages: base image from upstream source, then final image with Foundry tools
• Foundry version pinned in internal/openclaw/FOUNDRY_VERSION (currently v1.5.1)
• Daily cron checks for new upstream OpenClaw releases and rebuilds automatically

OpenClaw CLI Version Pinning

obolup.sh now pins and version-checks the OpenClaw CLI (npm package), matching the pattern used for kubectl, helm, k3d, and other binaries. Upgrades and downgrades are handled automatically during reinstall.

Other Changes

• macOS startup improvements (#210)
• OpenClaw bumped to v2026.2.23 with Renovate versioning fixes (#213)
• Frontend bumped to v0.1.9
• 21 embedded skills (up from 20) — added ethereum-local-wallet

Upgrading

bash <(curl -s https://stack.obol.org)
obol stack up
obol openclaw sync

Existing OpenClaw instances will have a wallet generated automatically on next sync.

Full Changelog: v0.3.2...v0.4.0

Release v0.4.0-rc0

Run OBOL_RELEASE=v0.4.0-rc0 bash <(curl -s https://stack.obol.org) to try this release candidate.

What's Changed

• Building a new openclaw image by @OisinKyne in #211
• chore(openclaw): bump to v2026.2.23 and fix Renovate versioning by @bussyjd in #213
• Local wallet by @OisinKyne in #212

Full Changelog: v0.3.2...v0.4.0-rc0

v0.3.2 - Reading Ethereum

v0.3.2 — Reading Ethereum

This release (improves stability and) updates the built in ethskills.com to get the most out of the Obol Stack's RPC infrastructure.

Get started with the quickstart guide here.

Highlights

Fix macOS cluster startup — k3d generates kubeconfig with server: https://0.0.0.0:<port> which fails on macOS with "can't assign requested address". The kubeconfig is now rewritten to 127.0.0.1 before saving, and obol stack up polls the k3s API server for readiness (up to 60s) before handing off to helmfile.

Resource limits across infrastructure — Previously, several infrastructure services ran with no resource bounds. All services now have explicit requests and limits to prevent resource starvation on local machines.

Dependency bumps — kube-prometheus-stack upgraded from 79.5.0 to 82.2.1 (Prometheus Operator v0.87 → v0.89), and obol-frontend bumped to v0.1.8.

OpenClaw Skills (20 built-in)

OpenClaw agents ship with 20 embedded skills from ethskills.com covering infrastructure operations, Ethereum development, and frontend development. These skills allow Openclaw agents to interact with the Ethereum mainnet and hoodi testnet. Upcoming updates will further improve their ability to use the chain to its fullest extent.

Infrastructure — ethereum-networks (read-only RPC queries via Obol's RPC), obol-stack (Kubernetes cluster diagnostics via ServiceAccount), distributed-validators (DVT cluster monitoring via Obol API)

Ethereum Development — addresses (verified contract addresses across chains), building-blocks (DeFi composability with Uniswap/Aave/Pendle), concepts (mental models for onchain building), gas (gas economics and L2 fee structures), indexing (The Graph and event indexing), l2s (Arbitrum/Optimism/Base/zkSync landscape), orchestration (end-to-end dApp build pipeline), security (vulnerability patterns and audit checklist), standards (ERC-20/721/4337/8004 specs), testing (Foundry unit/fuzz/fork/invariant tests), tools (Foundry/Hardhat/Scaffold-ETH 2), wallets (EOAs/multisig/AA/key safety), ship (idea-to-production routing), why (Ethereum + AI agent angle)

Frontend & UX — frontend-playbook (IPFS/Vercel/ENS deployment), frontend-ux (Web3 UX patterns for wallet flows), qa (pre-ship audit checklist)

Important

All of these features are in varying levels of alpha or beta, please report/pull request any issues encountered.

What's Changed

Fixes

• Fix macOS "can't assign requested address" on obol stack up by rewriting 0.0.0.0 → 127.0.0.1 in kubeconfig
• Add API server readiness poll after cluster creation to prevent helmfile race condition

Improvements

• Bump kube-prometheus-stack 79.5.0 → 82.2.1
• Bump obol-frontend v0.1.7 → v0.1.8
• Add resource limits to previously unbounded services:

Component	Requests	Limits
eRPC	100m CPU, 256Mi	1000m CPU, 2Gi
kube-state-metrics	10m CPU, 32Mi	200m CPU, 256Mi
node-exporter	10m CPU, 32Mi	200m CPU, 128Mi
llmspy	50m CPU, 64Mi	500m CPU, 512Mi

Full Changelog: v0.3.1...v0.3.2

What's Changed

• chore: bump obol-frontend to v0.1.7 by @agaskrobot in #203
• Change for read skills by @OisinKyne in #208
• Drop chart version by @OisinKyne in #209
• Startup not happy on mac by @OisinKyne in #210

Full Changelog: v0.3.1...v0.3.2

Release v0.3.2-rc1

Run OBOL_RELEASE=v0.3.2-rc1 bash <(curl -s https://stack.obol.org) to try this release candidate.

What's Changed

• Startup not happy on mac by @OisinKyne in #210

Full Changelog: v0.3.2-rc0...v0.3.2-rc1

Release v0.3.2-rc0

What's Changed

• chore: bump obol-frontend to v0.1.7 by @agaskrobot in #203
• Change for read skills by @OisinKyne in #208
• Drop chart version by @OisinKyne in #209

Full Changelog: v0.3.1...v0.3.2-rc0

Release v0.3.1

 

v0.3.1 — Obol Agents gain Obol Skills

This release pre-loads key skills for Agents running in the Obol Stack.

Get started on levelling up your Openclaw agents with the quickstart guide here.

Highlights

OpenClaw Agent Integration — Deploy and manage OpenClaw AI agents directly in the stack. A new obol openclaw command group provides setup, onboarding, and dashboard access. Agents are configured with your choice of model provider (Local Ollama, Anthropic, or OpenAI) and deployed as first-class Kubernetes workloads.

Obol Model Gateway (llmspy) — A cluster-wide OpenAI-compatible proxy that routes all LLM traffic. Cloud provider API keys are stored as Kubernetes secrets and managed via obol model setup. Supports provider waterfall routing and hot-swapping between Ollama, Anthropic, and OpenAI.

Application Management — Install arbitrary Helm charts as managed applications with obol app install, sync, list, and delete. Supports ArtifactHub references (bitnami/redis, bitnami/postgresql@15.0.0) with auto-generated helmfile configuration and petname namespaces.

Traefik + Gateway API — Replaced nginx-ingress with Traefik 38.0.2 using the Kubernetes Gateway API for HTTP routing. Cleaner routing rules for the frontend, ERPC, and per-network endpoints.

Cloudflare Tunnel Integration — Expose your stack to the internet via obol tunnel. Supports quick tunnel mode (default), persistent hostnames via obol tunnel login, and API-based provisioning.

Important

All of these features are in varying levels of alpha or beta, please report/pull request any issues encountered.

Contributors

• @bussyjd
• @OisinKyne
• @agaskrobot

What's Changed

• Fix default Ollama overlay missing api field for llmspy routing by @bussyjd in #181
• Release/pre flight and testing by @OisinKyne in #191
• Ability to update the stack and its apps gracefully by @OisinKyne in #179
• Add eRPC metadata ConfigMap for frontend discovery by @OisinKyne in #192
• fix(deps): bump golang.org/x/crypto to v0.45.0 by @bussyjd in #193
• fix: prevent k3d and k3s backends running side by side by @bussyjd in #194
• chore: bump obol-frontend to v0.1.6 by @agaskrobot in #200
• Openclaw skills by @OisinKyne in #197

Full Changelog: v0.3.0...v0.3.1

v0.3.0 - Agents

 

v0.3.0 — Obol Agents build with the Stack

This release brings AI agent infrastructure to the Obol Stack for the first time. OpenClaw agents can now be deployed and managed within the stack. These agents will be loaded with skills to deploy and manage cloud-native infrastructure alongside them in the stack.

Get started on levelling up your Openclaw agents with the quickstart guide here.

Highlights

OpenClaw Agent Integration — Deploy and manage OpenClaw AI agents directly in the stack. A new obol openclaw command group provides setup, onboarding, and dashboard access. Agents are configured with your choice of model provider (Local Ollama, Anthropic, or OpenAI. OAuth flows coming soon) and deployed as first-class Kubernetes workloads.

Obol Model Gateway (llmspy) — A cluster-wide OpenAI-compatible proxy that routes all LLM traffic. Cloud provider API keys are stored as Kubernetes secrets and managed via obol model setup. Supports provider waterfall routing and hot-swapping between Ollama, Anthropic, and OpenAI.

x402 Pay-per-Inference Gateway — Expose inference endpoints to the public internet with x402 USDC micropayment gating via obol inference serve.

Application Management — Install arbitrary Helm charts as managed applications with obol app install, sync, list, and delete. Supports ArtifactHub references (bitnami/redis, bitnami/postgresql@15.0.0) with auto-generated helmfile configuration and petname namespaces.

Traefik + Gateway API — Replaced nginx-ingress with Traefik 38.0.2 using the Kubernetes Gateway API for HTTP routing. Cleaner routing rules for the frontend, ERPC, and per-network endpoints.

Cloudflare Tunnel Integration — Expose your stack to the internet via obol tunnel. Supports quick tunnel mode (default), persistent hostnames via obol tunnel login, and API-based provisioning.

Important

All of these features are in varying levels of alpha or beta, please report/pull request any issues encountered.

Networks

• Networks refactor: Two-stage templating system (values.yaml.gotmpl → helmfile.yaml) with annotation-driven CLI flag generation, YAML validation, and deterministic field ordering
• Hoodi replaces Holesky as the default Ethereum testnet
• Helmfile now uses remote chart URLs and ArtifactHub resolution instead of bundled local charts

Infrastructure

• Upgraded pinned dependencies: kubectl 1.35.0, helm 3.19.4, k3d 5.8.3, helmfile 1.2.3, k9s 0.50.18
• Agent RBAC improvements and stdin support for obol agent init
• Obol agent merged into base infrastructure chart
• Frontend pinned to v0.1.4 with Renovate auto-bump configuration
• GitHub Actions pinned to commit SHAs for supply-chain security
• Bootstrap poll interval increased to 15 minutes
• Installer fixes for non-interactive mode and PATH handling

Breaking Changes

• Ingress controller changed from nginx to Traefik — existing clusters should be purged and re-created (obol stack purge -f && obol stack init && obol stack up)
• Network configuration directory structure changed by the networks refactor — previously installed networks should be re-installed

Contributors

• @bussyjd
• @OisinKyne
• @Padraic-O-Mhuiris
• @agaskrobot

Full Changelog: v0.2.0...v0.3.0

What's Changed

• Add Google API key secret management for Obol Agent by @bussyjd in #86
• Feature/obol agent refinements by @bussyjd in #119
• Created app install/sync/list/delete flow by @Padraic-O-Mhuiris in #118
• Renovate frontend version management by @bussyjd in #122
• Obol-agent template by @bussyjd in #84
• chore: upgrade pinned dependency versions in obolup.sh by @bussyjd in #123
• chore(frontend): pin image tag to v0.1.3 by @agaskrobot in #152
• chore(frontend): pin image tag to v0.1.4 by @agaskrobot in #154
• obolclaw by @OisinKyne in #160
• Tighten actions by @OisinKyne in #175
• Update readme and claude.md by @OisinKyne in #176
• Route around bad command by @OisinKyne in #178