Runtime-Agnostic Architecture

The @cloudflare/codemode package has been completely rewritten to support a modular, runtime-agnostic design. Code Mode continues to enable LLMs to write and execute code that orchestrates your tools, rather than calling them sequentially, resulting in significant token savings and improved model performance on complex tasks.

Key Improvements

New Executor Interface

• Minimal contract for code execution: execute(code, fns)
• Implement your own executor for any code sandboxing primitive or runtime
• Decouples the SDK from specific runtime dependencies

createCodeTool() Function

• Returns a standard AI SDK Tool for use in AI agents
• New import path: @cloudflare/codemode/ai
• Replaces the previous experimental_codemode() API

DynamicWorkerExecutor

• Runs generated code in Dynamic Workers
• Network isolation: fetch() and connect() blocked by default
• Console capture: Logs from console.log/warn/error returned in ExecuteResult.logs
• Configurable execution timeout (default 30 seconds)

Migration Guide

This release includes breaking changes:

• experimental_codemode() and CodeModeProxy have been removed
• The package no longer owns LLM calls or model selection
• Update imports to use createCodeTool() from @cloudflare/codemode/ai

To upgrade: npm i @cloudflare/codemode@latest

Refer to the Code Mode documentation for complete API reference and additional examples.