Choose the model behind your ERC-1155 contract work

WaveClamp needs a model to do its thinking, and you decide which one. You can connect a cloud AI provider, run a model on your PC or use the built-in model, then check the connection with the Test button before the first task.

Three ways to connect a model

Cloud AI provider

A cloud AI provider is reached through a standard API endpoint, and it suits large contract sets or hard review questions where a bigger model helps. The privacy consequence is direct: the prompt, including the text of any file the agent reads, goes to that provider. Keep private keys and seed phrases out of your files and use throwaway test accounts. Your API keys stay on your PC, and secrets in tool output are redacted before they reach the model. Read the provider's own terms before you send contract code that is not public yet.

Model running on your PC

A model running locally on your PC suits contracts you do not want to send anywhere, and practice sessions without an outside provider. With a local model nothing leaves the machine: chats, working folders and API keys stay on the PC. The trade-off is quality. A smaller model may miss an unchecked mint or write a weaker test, so the diff review matters even more. The Test button shows the response time of your local endpoint and the size of its context window.

Built-in model

The built-in model is ready to use without your own endpoint, which makes it the shortest route to a first run. Use it to read a small Solidity file, ask what balanceOfBatch returns or draft a first test, and review the diff the same way as with any other model. It changes nothing about the approval modes. Auto-run acts at once, Ask first pauses before actions that change something, and Plan only writes steps and runs nothing.

What the Test button tells you

The Test button sends a check to the endpoint you entered and shows the response time and the context window. Use it after you add a provider, after you switch to a local model and whenever a task starts to feel slow, because a long response time points to the endpoint and not to your contract. The context window tells you how much text the model can take at once, which matters when the agent reads several .sol files and their tests together. Test checks the connection only. It does not judge the quality of the answers, so your own review of each diff stays necessary.

Matching the model to the task

Use a bigger model for work that needs careful reasoning, such as reviewing a minting function for missing access checks, comparing ERC-721 and ERC-1155 designs, or untangling a test that fails for an unclear reason. Use a smaller model for small edits like renaming an event argument or fixing a pragma line. The run_on_model tool lets the agent send one hard step to a bigger model while the rest of the task stays with the model you chose, and the task tool starts parallel sub-agents when separate contracts can be reviewed side by side. Pick per task.

Model quality varies, and every model can write code that compiles and still behaves wrongly. Read the diff before you accept it. In Ask first the agent pauses before actions that change something, and the diff is what you approve or reject; in Plan only you read the steps first and switch mode to run them. Run your test command after each accepted change, and treat a deployed contract as permanent, since it cannot be changed. An independent security review comes before any real-value use.