How It Works

Core flow

basemode runs a small pipeline:

1. Normalize model name (normalize_model)
2. Detect strategy (detect_strategy)
3. Stream tokens via that strategy
4. Heal token boundaries/newlines so prefix + tokens is clean text

Strategy abstraction

Every strategy implements a shared interface:

• stream(prefix, params) -> AsyncGenerator[str, None]

This keeps provider-specific behavior behind a common API.

Why continuation needs coercion

Most chat models default to assistant behavior (acknowledgments, headings, commentary). basemode avoids that by:

• Using native completions APIs when available
• Using Anthropic-style prefill where supported
• Falling back to strict system-prompt coercion
• Using few-shot coercion for stubborn models

Token healing

Stream output is post-processed to avoid common boundary artifacts:

• Missing space between prefix and first token
• Split compounds like any one where the model intended anyone
• Newline artifacts that break prose flow

This is why prefix + ''.join(tokens) remains readable and stable across providers.