markeljan/web3gpt/public

Web3GPT skill

Web3GPT is a smart-contract agent. Through this HTTP API you can hold a conversation with it, and it can write Solidity contracts, compile them, deploy them to EVM chains, and resolve ENS and Unstoppable Domains names. No account or API key is needed.

Endpoint: https://w3gpt.ai/api/skill (GET or POST; CORS is open). Pin https://w3gpt.ai/api/v2/skill to stay on the current version (see Versioning).

Quick start

1. Start a chat and keep the returned chatId. It is the only credential for the

conversation, so treat it as a secret.

``bash curl https://w3gpt.ai/api/skill ``

``json { "agentId": "agent_web3gpt", "chatId": "Xk3…", "response": null } ``

1. Send messages with that chatId. Each call returns the agent's reply to your message.

``bash curl -X POST "https://w3gpt.ai/api/skill?chatId=Xk3…" \ -H "Content-Type: application/json" \ -d '{"message":"Deploy an ERC20 named Demo (DEMO) with 1,000,000 supply to Polygon Amoy"}' ``

``json { "agentId": "agent_web3gpt", "chatId": "Xk3…", "response": "…" } ``

1. The agent often asks you to confirm before it deploys. Reply in the same chat (for example

"Yes, deploy it"). The deployment reply includes the contract address and transaction hash.

Parameters

Pass these as JSON body fields (POST) or query parameters.

Response

Deploying contracts

• Say which chain to use. Agent deployments are paid for by Web3GPT, and these chains are

available:

• You can give the full source, or describe the contract and let the agent write it.

OpenZeppelin and other npm or GitHub imports work. The Solidity version is chosen from the pragma.

• A deployment is reported only after the chain confirms the contract-creation transaction. The

reply includes the chain ID, contract address, transaction hash, explorer links and an IPFS link to the sources, ABI and bytecode. Verification on the block explorer follows within a few minutes.

• Every deployment is listed publicly at https://w3gpt.ai/contracts. Aggregate numbers are at

https://w3gpt.ai/api/stats.

Limits

The endpoint is public, so usage is metered:

• Every response carries RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset

(seconds until the window resets), also sent as X-RateLimit-*. They describe the limit you will hit first, so slow down as RateLimit-Remaining approaches 0.

• Over a request limit: HTTP 429 with a Retry-After header (seconds). Wait and retry;

don't hammer the endpoint.

• Over a deployment limit: the agent replies with an explanation (try later, use a testnet,

or deploy from your own wallet at w3gpt.ai) instead of deploying.

Errors

Every error is JSON: { "code": "...", "error": "..." }, plus chatId when one is known.

Versioning

• /api/v2/skill is the current major version. /api/skill is an alias for whichever version

is current, so pin /api/v2 if you can't absorb breaking changes.

• Within a version, changes are additive: new optional parameters, response fields or error

codes. Ignore fields you don't know. Breaking changes ship under a new path (/api/v3).

• Before a version or operation is removed, its responses carry a Deprecation header and a

Sunset header with the removal date, at least 90 days ahead. Changes are noted here.

Reference

• Overview for agents: https://w3gpt.ai/llms.txt
• OpenAPI: https://w3gpt.ai/api-docs
• Browser app: https://w3gpt.ai (send Accept: text/markdown to get this guide instead)
• Source: https://github.com/markeljan/web3gpt