Unified token usage across all providers. Log line example: [TOKEN USAGE] provider=gemini model=... | input=1000 output=500 c

Per-million-token pricing in USD.

An image a tool wants the model to see. Exactly one source is set: ``data`` for local images (sent as bytes to Gemini / base64 to Anthropic &

Handles selection of variant models based on available API keys and generation type

Handles parameter extraction and validation from WebSocket requests

Handles WebSocket communication with consistent error handling

Writes one JSON report per LLM API request (turn). ``record_request`` writes the report before the request is sent so a report exists even if

Context object that carries state through the pipeline

Sends initial status messages to all variants

Handles agent tool-calling generation for each variant.

Handles the main code generation logic

Handles parameter extraction and validation

Pipeline for processing WebSocket code generation requests

Default backend: renders in local headless Chromium. Runs locally, so the page can load assets served from localhost (e.g. /local-assets/ URL

Handles post-processing and logging

Handles post-processing after code generation completes

Handles prompt creation

Handles prompt assembly for code generation

Handles WebSocket setup and teardown