AI Meeting Summarizer: Free Local Pipeline with Whisper and Ollama
An AI meeting summarizer you can run free and local: Whisper for transcription, an LLM for summaries, n8n for the pipeline. Real setup included.
💡 What You Will Learn
An AI meeting summarizer you can run free and local: Whisper for transcription, an LLM for summaries, n8n for the pipeline. Real setup included.
Meeting summaries are the perfect self-hosted AI project: the audio is private, the task is boring, and the pipeline is simple enough to build in an afternoon.
The Pipeline
Whisper (106,813 stars) transcribes the recording; faster-whisper (24,790 stars) does the same job 4x faster with CTranslate2 for long meetings. The transcript then goes to a local LLM via Ollama (177,874 stars) with a summary prompt: decisions, action items, owners, deadlines. n8n (199,484 stars) ties it together - watch a folder, transcribe, summarize, post to Slack or Notion.
Quality Notes
Whisper large-v3 handles most languages well; speaker labels need whisperX (23,466 stars). For action items, prompt the LLM to extract them in a strict list format - vague summaries are useless summaries. Privacy win: a 2-hour board meeting never leaves your server.
FAQ
Q: What hardware do I need?
A: faster-whisper on CPU handles a 1-hour meeting in about 10 minutes; a GPU cuts that to 2-3 minutes.
Q: How accurate is it?
A: Clear audio hits 90%+ word accuracy; accented or overlapping speech drops it - keep the transcript, do not trust it blindly.
