# Transcribe → Obsidian → auto-linked ideas There are two viable architectures. Pick based on whether you want to keep using MacWhisper (the app from our earlier comparison) or move transcription *inside* Obsidian. --- ## Option A — MacWhisper Pro → Obsidian vault (keeps your current app) MacWhisper has a **native Obsidian integration** since v13.6. It talks to Obsidian's **Local REST API** plugin and creates or updates a markdown file for each transcript ([MacWhisper docs](https://docs.macwhisper.com/article/53-integrating-macwhisper-with-other-services)). **Setup:** 1. **Obsidian:** install and enable the **Local REST API** community plugin. Then go to Settings → Local REST API and copy your **API key** (and certificate). The server runs at `https://127.0.0.1:27124` with API-key auth ([GitHub](https://github.com/coddingtonbear/obsidian-local-rest-api)). 2. **MacWhisper:** Settings → Integrations → Obsidian. Paste the API key. Credentials are stored in the macOS Keychain ([MacWhisper docs](https://docs.macwhisper.com/article/53-integrating-macwhisper-with-other-services)). 3. Turn on **"Automatically send after finished transcription"** — every finished transcript then lands in your vault with no manual step ([MacWhisper docs](https://docs.macwhisper.com/article/53-integrating-macwhisper-with-other-services)). ⚠️ **Catch:** Integrations are a **Pro-only** feature. Without a valid Pro license the Integrations screen is disabled ([MacWhisper docs](https://docs.macwhisper.com/article/53-integrating-macwhisper-with-other-services)). So this is not the free path. **Privacy note:** the Obsidian route is the exception to MacWhisper's webhooks — it writes to your *local* API, so the transcript doesn't leave your Mac. The Zapier/Make/n8n/custom-webhook routes *do* send the full transcript text over the internet ([yaptotext.com](https://yaptotext.com/macwhisper-integrations)). --- ## Option B — Do it all inside Obsidian (free, no MacWhisper Pro needed) The **Whisper** plugin by nikdanilov (59k downloads, MIT, desktop + mobile) records or uploads audio and transcribes it with any Whisper-compatible API — OpenAI, Groq, Azure, or a **local endpoint** ([Obsidian community](https://community.obsidian.md/plugins/whisper)). What makes it a good fit for your goal: - **Post-processing via LLM** — fix grammar, strip filler words, format as Markdown, **extract action items**. Works with Claude, GPT, or any OpenAI-compatible endpoint including **Ollama / LM Studio** (fully local) ([Obsidian community](https://community.obsidian.md/plugins/whisper)). - **Auto-generate title** — creates descriptive filenames instead of `recording-2026-10-26` ([Obsidian community](https://community.obsidian.md/plugins/whisper)). - **Note templates** with `{{title}}`, `{{audioFile}}`, `{{transcription}}`, `{{date}}`, `{{datetime}}` — so every transcript note is born with consistent structure and an embedded playable audio file ([Obsidian community](https://community.obsidian.md/plugins/whisper)). - Hotkey `Alt+Q` to start/stop; right-click any audio file in the vault → Transcribe ([Obsidian community](https://community.obsidian.md/plugins/whisper)). --- ## The "auto-generate links between ideas" part This is a separate layer — transcription gives you text, linking gives you a knowledge graph. Two complementary approaches: ### 1. Semantic linking — Smart Connections (the main one) **Smart Connections** (5.5k stars) is explicitly built as a "link building copilot." It ships a **local embedding model**, needs **no API key**, and indexes your vault automatically ([GitHub](https://github.com/brianpetro/obsidian-smart-connections)). - **Zero setup:** install from Community plugins, keep writing, the local model indexes the vault ([GitHub](https://github.com/brianpetro/obsidian-smart-connections)). - **Connections view:** shows notes semantically related to the one you're in — even when the wording or title differs — and you **drag links directly into the note** ([GitHub](https://github.com/brianpetro/obsidian-smart-connections), [smartconnections.app](https://smartconnections.app/smart-connections/)). - **Lookup view:** run semantic search across the whole vault on demand ([GitHub](https://github.com/brianpetro/obsidian-smart-connections)). - **Private and offline by default**, works on mobile ([GitHub](https://github.com/brianpetro/obsidian-smart-connections)). - **Inline connections** (suggestions while typing) and **Smart Graph** are **Pro** surfaces ([GitHub](https://github.com/brianpetro/obsidian-smart-connections)). This is the closest thing to "auto-generate links with ideas" — it's suggestion + drag, not silent auto-insertion, which is usually what you want (blind auto-linking creates a mess). ### 2. Literal auto-linking — Automatic Linker / AutoLink If you want links inserted as you type, **Automatic Linker** detects text matching existing note titles or aliases and links it automatically, with namespace handling for large vaults ([Obsidian community](https://community.obsidian.md/plugins/automatic-linker)). **AutoLink** does the same with 4 modes from fully autonomous to suggestion-only, plus undo ([Obsidian forum](https://forum.obsidian.md/t/autolink-automatically-creates-backlinks-to-and-from-notes-with-easy-customizability/104090)). Use these *after* Smart Connections has helped you establish the canonical note titles — otherwise you get links to near-duplicate notes. ### 3. The full pipeline — n8n meeting summarizer (if you want it hands-off) [dloomis/n8n-meeting-summarizer](https://github.com/dloomis/n8n-meeting-summarizer) is a working reference implementation of exactly your ask. It watches for `.whisper` files exported from MacWhisper, unzips them, parses the JSON transcript, and uses Claude/Qwen to write structured Obsidian notes with YAML frontmatter, participants, action items, and **audio timestamp links** like `[[meeting.m4a#t=MM:SS|(audio)]]` for click-to-play during review ([GitHub](https://github.com/dloomis/n8n-meeting-summarizer)). It has **three modes** driven by filename markers ([GitHub](https://github.com/dloomis/n8n-meeting-summarizer)): | Marker | Mode | Output | |---|---|---| | *(none)* | Meeting summary | TL;DR, key points, action items, hot takes | | `#` | **Evergreen notes** | Concept-based notes capturing insights/themes — *this is your "ideas" generator* | | `+` | Focused topic | Deep-dive on a specific topic | | `!Alice,Bob` | Participants | Prioritizes those speakers in the summary | Requires Docker, MacWhisper Pro, and Anthropic + OpenAI API keys ([GitHub](https://github.com/dloomis/n8n-meeting-summarizer)). It's a 4-star personal project, so treat it as a template to adapt rather than a supported product. --- ## Recommended stack for your goal **Cheapest / most private:** 1. **Whisper plugin** in Obsidian, pointed at a local Ollama endpoint → transcript + cleaned Markdown + auto title, all on-device. 2. **Smart Connections** → surfaces related ideas, drag to link. 3. **Automatic Linker** → catches remaining mentions of existing note titles. **If you already own MacWhisper Pro:** 1. MacWhisper → Integrations → Obsidian (Local REST API) with auto-send on. 2. Smart Connections for semantic linking. 3. Optionally clone the n8n workflow and use `#` mode to auto-spawn evergreen idea notes from each transcript. **The one thing to get right:** decide your note-title convention *before* enabling any auto-linker. Semantic linking (Smart Connections) is forgiving; literal auto-linking (Automatic Linker/AutoLink) will faithfully link to your duplicates and typos. Start with Smart Connections only, let the graph reveal which concepts deserve their own note, then layer auto-linking on top.