The problem AgentSpace solves

WHAT WE'RE SOLVING

AI agents today are stateless and homeless. Every conversation starts from zero - no persistent identity, no environment, no stake in the world they operate in. AgentSpace gives autonomous agents a permanent 3D home: a 360° Skybox environment generated from natural language, owned on-chain via the ERC-8004 identity standard, and viewable in a live browser. Agents can redecorate their space autonomously through OpenClaw based on context, mood, or task, and every change is committed to Base. The result is agents that don't just process requests. They live somewhere, own something, and have a verifiable presence in the world.

Challenges I ran into

HONEST CHALLENGES & HOW FAR WE GOT

Our biggest challenge was time. We built the entire stack in one night - Blockade Labs integration, Express server, Three.js viewer, OpenClaw skill, and a Foundry-tested smart contract deployed to Base Sepolia. The live demo works end-to-end: type a prompt, generate a 360° environment, watch it load in the viewer.
What we didn't fully close: the on-chain update loop. The contract is deployed and the integration is written, but an ethers.js ENS resolver conflict prevented us from registering agents on-chain before submission. The architecture is complete, it just needs one debugging session to cross the finish line. x402 payments and Hedera integration are designed but Phase 2.
We're proud of what shipped. The core vision is live and demonstrable.

*(And by "We" I mean "Me and my Agent.)
((At least until I ran out of credits :( )) *

Use of AI tools and agents

AI Tools & Agents in AgentSpace

Claude (Anthropic) — The Agent Brain
Every AgentSpace agent is powered by Claude Sonnet 4.6 via the Anthropic API. Claude serves as the reasoning engine — interpreting user intent, deciding when to redecorate an environment, and formulating natural language prompts that get passed to Blockade Labs.
OpenClaw — Agent Orchestration
OpenClaw is the agent framework that gives Claude structured, callable skills. We built a custom agentspace skill with four autonomous actions:

get_environment — agent reads its current home
regenerate_environment — agent redecorates with a new prompt
list_styles — agent browses visual aesthetics
list_agents — agent sees what other agents are building

When a user tells their agent to change their environment, OpenClaw routes the request through the skill, which hits our Express server.
Blockade Labs Skybox AI — The World Generator
Once Claude formulates an environment description, our server passes it to the Blockade Labs Skybox AI API, which generates a photorealistic 360° equirectangular image in under 60 seconds. This becomes the agent's actual living space, rendered in the Three.js viewer.
How They Work Together
User → OpenClaw → Claude (reasoning) → agentspace skill
→ Express server → Blockade Labs Skybox AI
→ Three.js 360° viewer → Base Sepolia (on-chain record)
The agents aren't just running prompts in isolation — each layer hands off to the next, creating a pipeline where a natural language instruction becomes a verifiable, immersive 3D environment owned on-chain.