Local Code AI VS Code Extension

This extension is an extension-only VS Code client for the Local Code AI structured stdio bridge. It does not include an agent executable, model weights, or an inference runtime. It does not provide a second inference engine or send workspace content to a cloud service.

Installing the extension alone is not sufficient for chat or offline inference. Before use, the user must independently obtain a compatible local bridge executable, a permitted GGUF model, and a compatible llama.cpp runtime. Each asset has separate licensing and provenance requirements. Do not redistribute model or runtime assets unless their exact terms permit it.

Portable package prerequisites

For the current Windows package, the user must supply and configure all of the following separately:

• LocalCodeAI.exe from an authorized application distribution.
• A GGUF model from a source whose terms permit the intended use.
• A compatible Windows llama.cpp executable and all required sibling DLLs.

The extension does not download, install, or select any of these assets.

For a source checkout, configure:

{
  "localCodeAI.agentExecutable": "C:\\path\\to\\local-code-ai\\.venv\\Scripts\\python.exe",
  "localCodeAI.agentArgs": ["-m", "app.main", "--bridge"],
  "localCodeAI.agentRoot": "C:\\path\\to\\local-code-ai"
}

For the portable package, use LocalCodeAI.exe and --bridge; set localCodeAI.agentRoot to the package directory if VS Code launches elsewhere. The bridge uses a private newline-delimited JSON stdin/stdout pipe and opens no TCP listener.

Safeguards include bounded retrieval, sensitive-file exclusions, exact patch proposals opened in the standard VS Code diff editor before explicit approval, workspace-constrained paths, and no automatic model-triggered shell execution. The command field requires a separate modal user approval and runs with shell=False, a workspace-relative working directory, a 120-second timeout, and bounded output.

The extension can be packaged without a project dependency install with:

npm exec --yes @vscode/vsce -- package --no-dependencies

This produces local-code-ai-0.1.0.vsix in the extension directory.

Before Marketplace publication, replace the placeholder publisher value in package.json with the publisher account that owns the listing. Publish the extension separately from any application, model, or runtime distribution.