0G Thermidora

The Vision

We set out to build an AI Agents platform that could interact with the 0G blockchain - agents that upload files to decentralized storage, monitor chain activity, manage key-value data, and alert on security anomalies. Agents that understand the 0G network natively and respond with real mainnet data. We wanted to make the platform accessible by providing readily available agent skills and templates that integates with 0G so builders can get started easy.

But we quickly hit a wall. There was no managed infrastructure for running persistent AI agents that fit our needs. Every existing option meant self-hosting — setting up servers, configuring Docker, managing networking, TLS, storage, process restarts, and updates. All before writing a single line of agent logic. The operational overhead was the bottleneck, not the AI.

So we built the core first.

Thermidora - The core we had to build

Thermidora is a fully managed API platform for deploying, hosting, and orchestrating AI agents. It abstracts away everything — provisioning, compute isolation, persistent storage, networking — and exposes it all as a simple REST API. No servers. No Docker. No ops. Just HTTP calls.

It's built on OpenClaw, an open-source AI agent framework that gives agents memory, tool use, skills, cron scheduling, browser control, and multi-channel messaging (Telegram, Discord, Slack, WhatsApp, and more). OpenClaw is powerful - but self-hosting it in production is a project in itself, especially if we want to make all interactions API-first. Thermidora wraps OpenClaw in managed infrastructure so developers can skip the ops and go straight to building.

Every operation in Thermidora is an API call. You can go from zero to a running AI agent in under three minutes with nothing but API calls.

Live at: [thermidora.com](https://thermidora.com) · API Docs: [thermidora.com/docs](https://thermidora.com/docs)

0G Thermidora - What we built on top

With Thermidora as the foundation, we built 0G Thermidora — an AI agent platform purpose-built for the 0G blockchain network. It provides pre-built agent templates and skills, each specialized for a different aspect of 0G operations, all connected to 0G Mainnet (Aristotle, chain 16661) and responding with real blockchain data.

Live at: [0g.thermidora.com](https://0g.thermidora.com)

Agent Templates

• StorageBot — Upload, download, pin, and monitor files on 0G decentralized storage. Tracks quotas, costs, and expiring files. Uses the 0g-storage skill.

• ChainWatch — Monitors 0G Mainnet activity like whale alerts, gas price tracking, block queries, and transaction analysis. Uses the 0g-chain skill.

• KVault — Manages key-value data on the 0G network like read/write operations, batch processing, state snapshots, and schema management. Uses the 0g-kv skill.

• Sentinel — Security-focused monitoring like contract audit summaries, anomaly detection, suspicious activity alerts, and access control analysis. Uses the 0g-chain skill.

0G Skills

Skills are modular capabilities installed on Thermidora instances that give agents blockchain powers:

• 0g-storage — Upload, download, pin, unpin, list files, check storage usage on 0G decentralized storage

• 0g-chain — Get balances, send transactions, read contracts, query blocks, watch events, estimate gas on 0G Mainnet

• 0g-kv — Get, set, delete, list keys, and batch operations on 0G's decentralized KV store

How it works

The entire platform is API-driven. No SDKs to install, no CLI tools, no Docker to configure:

1. Create an instance — POST /v1/instances provisions an isolated OpenClaw runtime with persistent storage

2. Install 0G skills — POST /v1/instances/:id/skills adds blockchain capabilities to the instance

3. Deploy an agent — POST /v1/instances/:id/agents with a persona, model selection, and skill bindings

4. Send messages — POST /v1/instances/:id/agents/:agentId/messages with optional SSE streaming for real-time token-by-token responses

The Thermidora API - In detail

Reference to the API docs: https://www.thermidora.com/docs

The API is the product. Every aspect of agent lifecycle, communication, and management is a REST endpoint.

Architecture

Technical Stack

• 0G Thermidora Frontend and Demo App — Next.js 15, React 19, TypeScript

• 0G Thermidora API Routes — Next.js API routes with Thermidora Core proxy and SSE passthrough

• Thermidora Core API — AWS Lambda (Node.js/Express) behind CloudFront, with SSE support

• Thermidora Core Compute — AWS ECS Fargate (ARM64), fully isolated containers per instance

• Thermidora Core Storage — Amazon EFS (persistent agent workspaces), S3 (messages, events, files)

• Thermidora Core Database — DynamoDB (accounts, instances, agents, API keys, schedules, feedback)

• Thermidora Core Scheduling — EventBridge Scheduler for cron jobs and background agent tasks

• Thermidora Core Container Registry — Amazon ECR with thin-layer overlay images on OpenClaw base

• Agent Runtime — OpenClaw open-source AI agent framework (memory, tools, skills, sessions, multi-channel)

• AI Models — Amazon Bedrock Claude Sonnet 4/4.5/4.6, Kimi K2.5, Qwen 3 32B, DeepSeekR1

• Blockchain — 0G Mainnet (Aristotle, chain 16661)

• Deployment — AWS CDK (infrastructure), GitHub Actions (CI/CD)

Key Design Decisions

• API-first, always — Every operation is a REST call. The 0G showcase demonstrates what the API can do; the Deploy dialog shows the exact curl commands to replicate it. No SDKs, no CLI, no vendor lock-in.

• We built the platform first — To build great blockchain agents, we needed great agent infrastructure. Thermidora is that infrastructure — and it's general-purpose enough to power any AI agent use case, not just 0G.

• OpenClaw as the runtime — We didn't build an agent framework from scratch. OpenClaw is battle-tested, open-source, and gives agents memory, tool use, skills, cron, and multi-channel messaging out of the box. Thermidora manages it so developers don't have to.

• Isolation by default — Each Thermidora instance runs in its own ECS Fargate task with dedicated compute (up to 1 vCPU / 4 GB) and persistent EFS storage. No shared state between users. Agents are sandboxed.

• Skills as modular capabilities — 0G skills 0g-storage, 0g-chain, 0g-kv) are installed at the instance level. The same skill works on testnet or mainnet — network config lives in environment variables. Custom skills can be bundled and uploaded via the API.

• Transaction approval flow — For write operations that cost A0GI (uploads, pins), agents produce unsigned transaction parameters that surface as approval cards in the UI. The user's wallet (MetaMask) signs in the browser — agents never hold private keys.

This project was built entirely during the hackathon period. Both Thermidora (the core managed agent platform) and 0G Thermidora (the 0G integration layer, skills, agent templates and demo site) were developed from scratch starting March 21, 2026.

Development Timeline

• Week 1 - Foundation (March 21–28): Scaffolded the monorepo, mockup the initial dashboard UI, whiteboard the API interface and components, deep dive on OpenClaw implementations.

• Week 2,3 and 4 - Thermidora Core Development (March 28 – April 23): Built the full core Thermidora API backend on AWS (ECS Fargate instances, agent lifecycle, SSE message routing, skills, EventBridge cron, DynamoDB, API key + JWT auth), and researched the 0G SDK and network architecture. Deployed at [thermidora.com](https://thermidora.com).

• Week 4 and 5 - 0G Integration (April 19–26): Built the integration with 0G and a demo showcase at [0g.thermidora.com](https://0g.thermidora.com), developed 0G specific agent skills (0G Storage, 0G Chain, 0G KV Store), and shipped 4 live sample agents (StorageBot, ChainWatch, KVault, Sentinel) connected to 0G Mainnet, and built the "Try it out" and "Deploy" dialogs with live SSE streaming.

• Week 6 - Showcase & Polish (April 27 – May 9): Submission preparation, demo video, and final testing.

No fundraising at this time. The project is self-funded and supported by AWS credits which cover the initial infrastructure costs for running Thermidora 0G. This is sufficient to support the current scale and early adoption.