Skip to content

ADR-009: Privacy-First Local Summarization

  • Status: Accepted
  • Date: 2026-01-11
  • Authors: Podcast Scraper Team
  • Related RFCs: RFC-012
  • Related PRDs: PRD-005

Context & Problem Statement

Summarizing long podcast transcripts often involves sensitive or personal content. While Cloud LLMs (OpenAI, Anthropic) are easy to use, they introduce privacy risks, recurring costs, and internet dependencies.

Decision

The project defaults to Local Summarization using Transformer models (BART, LED) running on the user's hardware. Cloud-based summarization is treated as an optional secondary provider.

Rationale

  • Privacy: Transcripts never leave the user's machine.
  • Cost: Zero per-token cost allows for processing massive back-catalogs without financial overhead.
  • Reliability: The pipeline works offline and is immune to API rate limits or service outages.

Alternatives Considered

  1. Cloud-First: Rejected to maintain the "toolbox" nature of the project and respect user privacy.
  2. Rule-based Summarization: Rejected as it cannot capture the nuance and abstractive qualities of podcast conversations.

Consequences

  • Positive: High privacy; zero marginal cost.
  • Negative: Requires significant local hardware (RAM/GPU); summaries may have lower "polish" than premium cloud models.

References