Audio version — Estimated duration: 3 min 21 sec

”CEREBRAS_API_KEY not set”

You need the environment variable for online LLM mode.

$env:CEREBRAS_API_KEY = "your-key-here"

Or switch to local LLM: --llm local

”llama.cpp call failed”

Ensure the llama.cpp server is running:

llama-server -m your-model.gguf --port 8080

Check it responds:

curl http://localhost:8080/v1/models

“No .md files found”

The scanner skips hidden directories (names starting with .). Ensure your files are in a non-hidden directory.

”File too short or empty”

Files under 10 characters of content are skipped. This filters out stub files and empty templates.

Piper not downloading

The download URL is constructed from the voice name. Ensure:

  • The voice name is a valid Piper voice from Hugging Face
  • You have internet access
  • huggingface-hub or urllib can reach huggingface.co

Audio embed not appearing

The injector checks for the **Audio version** marker. If the file already has one, it is skipped to prevent duplicates. Remove the existing marker to re-inject.

”Could not read” errors

Check file permissions and encoding. The reader expects UTF-8.

Rate limit hit

Online LLM mode enforces 5 requests/minute. The pipeline waits automatically. For batch processing many files, consider local LLM mode.