Skip to content
IdeaScout.
← Back to archive

Local-first meeting transcription that pipes into Obsidian/PKM

AI-discovered

Problem

A developer of a popular Obsidian-Granola sync plugin was forced to deprecate it because Granola cracked down on auth access. In the 'cambrian explosion' of meeting transcription tools, he could not find one that runs as a background daemon, uses local models for transcription and diarization, is CLI-first with no mandatory cloud UI, and cleans/summarizes transcripts into his own knowledge base. Every existing option is a closed SaaS that locks your meeting notes inside their platform.

Opportunity

A privacy-first, local-model meeting transcription daemon (macOS/Windows) with calendar integration, automatic cleanup/summarization, and open export formats (Markdown/Obsidian, API endpoints). Targets the large PKM/Obsidian community that already pays for plugins and values data ownership.

Market analysis

The pain is authentic, but searches show this exact niche is already populated by free open-source projects: ownscribe is a local-first transcription-and-summarization CLI, OpenWhispr runs Whisper/Parakeet plus local pyannote diarization, and JoseEliel/scribe streams local LLM analysis through Ollama. What none of them fully nails is the daemon-plus-calendar-plus-Obsidian polish, but that gap is small enough to be closed by any of them rather than justifying a new product.

Market · Privacy-conscious developers and PKM/Obsidian users; real willingness to pay exists in the Obsidian plugin ecosystem, but the free OSS alternatives cap price expectations hard.

Pricing · Obsidian community norms: one-time licenses of $20-60 or $5-10/mo for premium plugins; competing free tools (noScribe, ownscribe, OpenWhispr) make a subscription pitch difficult unless calendar integration and summarization are visibly superior.

score 4/10 by glm-5.1

Pros

  • + Verified, current pain: a popular sync plugin was just deprecated over auth crackdowns.
  • + Clear differentiation spec from the source thread (daemon, CLI-first, local diarization).
  • + Obsidian/PKM users demonstrably pay for quality local tooling.

Cons

  • − Several active FOSS projects already cover most of the feature list.
  • − System-audio capture permissions (especially macOS TCC) are a support minefield.
  • − Local-model performance expectations on consumer laptops lead to chronic dissatisfaction.

Source

Hacker News (Ask HN)

Open original thread ↗

The uncomfortable truth from the search results is that the ‘no local-first option exists’ premise is already stale — the field erupted in the last year and the remaining gap is integration polish, not architecture. A new entrant’s only durable angle is being the best Obsidian citizen (templates, frontmatter, dataview-compatible metadata, calendar-joined file naming) rather than the best transcriber, because transcription quality is a commodity bounded by local Whisper-class models. Also note the second-order risk that made this idea topical in the first place: any product whose hook is syncing with a closed SaaS (Granola, Otter, Fireflies) inherits that vendor’s API whims — so the export-first, local-only stance is not just privacy theater, it is the moat.