Voice note processing — Monologue MCP integration plan (2026-05-10, 07:18)
Earliest of the 5/10 Monologue cluster — the design seed for what became the fetch-monologue skill. Names the now-implemented shape: drop the local
transcripts/folder, add a Phase-0 fetch inprocess-inboxthat compares Monologue notes against a local log, downloads full transcripts (not summaries), and routes them through the existing pipeline. Cross-platform Monologue use on Mac + iPhone is the input premise; the rest of the day’s frustration thread (2026-05-10-Voice memo workflow frustration - Monologue, Shortcuts, Working Copy, Zettel) is what made this design urgent.
Voice Transcript
Recorded: May 10, 2026 at 7:18 AM
Well, so anyway, let’s see if that gets, that gets used. So this is like the notes feature here where you can just talk. [clears throat] And it’s pretty nice because you can use it on your Mac and on the iPhone too. And so as a to-do, I will probably have to, uh, set it up in Cloud Code with the skills and stuff as well, but it should, it should be fine. Uh, so small refactoring and essentially removing the transcripts folder from the Obsidian vault, and then, well, integrating it so that on every process inbox run, there’s also a process monologue notes skill that basically fetches all the notes and compares to some… Probably makes sense to have some log file, uh, locally that says which notes have been processed already, and then basically for the ones that are new, it then, uh, looks at the transcripts, the full one, not the summary, and, uh, downloads it into the vault. I guess that makes sense, and also processes it. However, it’s like a new approach compared to where the transcripts are already inside the vault. But I guess the rest, like after actually downloading the content, should be the same as it is right now, so it shouldn’t change too much. But yeah.
What got implemented
The design here maps directly onto the current vault pipeline:
| Idea in this memo | Implemented as |
|---|---|
| ”set it up in Cloud Code with the skills” | fetch-monologue skill, callable as /fetch-monologue and run as Phase 0 of process-inbox. |
| ”small refactoring and essentially removing the transcripts folder” | 0_Inbox/transcripts/ is now a transient drop zone written by fetch-monologue, not a permanent vault folder of pre-existing transcripts. |
| ”process monologue notes skill that basically fetches all the notes and compares to some… log file” | 7_Agent/monologue-processed.md is the dedup log — keyed by monologue_id UUIDs, appended on each fetch, never pruned. |
| ”for the ones that are new, it then looks at the transcripts, the full one, not the summary, and downloads it into the vault” | fetch-monologue calls get_note with include_transcript: true and writes only the transcript body — Monologue’s summary is dropped (the user dislikes its style). |
| “the rest, like after actually downloading the content, should be the same as it is right now” | Routing, cross-linking, log entry, deletion — process-inbox Phase 1 unchanged. |
Related
- fetch-monologue — the skill this memo proposed; the implementation now sitting in
7_Agent/skills/. - process-inbox — the parent skill that runs
fetch-monologueas Phase 0; this memo is the design seed for that integration point. - AGENTS — the contract notes that voice memos live in Monologue and
fetch-monologueis Phase 0 ofprocess-inbox— the documented form of the plan articulated here. - 2026-05-10-Voice memo workflow frustration - Monologue, Shortcuts, Working Copy, Zettel — same-day later (14:00) rant about the cost of Monologue and the fragility of the previous Shortcuts-based rig; that frustration motivated the smoother MCP-based pipeline this memo prototypes.
- 2026-05-10-Inbox-style PKM app - input funnel and iOS sync challenge — same-day later (14:07) product-idea sibling: a standalone app version of the inbox funnel; this memo is the within-Obsidian version of the same input-funnel shape.
- 2026-05-10-Inbox routing for work notes - Birchill folder plan — same-morning follow-up (07:25): adds the work/private split as a downstream concern for the same Monologue → inbox pipeline.
- 2026-05-10-Apple Watch shortcut wrapper vs direct action — adjacent 07:19 jot about the still-needed Apple Watch capture surface for the same pipeline.
- 2026-03-24-Standalone voice note transcription app — six-weeks-earlier “build my own dictation app” pull; this memo is the cheaper “keep the dictation tool, write the integration plumbing” answer.
- 2026-03-24-Dictation feature for Zettel vs standalone app — sibling on the build-vs-buy question; settled in favour of buy + integrate, which is what this memo encodes.
- My AI native Obsidian Setup — the parent setup page; the Monologue-fetch loop here is now one of the pipeline’s input mouths.
- Idea – Inbox Prompt Types — adjacent idea about the inbox as an instruction channel; this memo is closer to the transcript-as-content channel for the same inbox.
- OpenClaw for PKM — the longer-arc PKM-as-app project; this memo stays inside Obsidian rather than reaching for the standalone-app answer, but the pattern is the same.
- 2026-05-21-Monologue auto-records on open — 11-days-later UX-property jot naming why Monologue is the right capture surface for this pipeline: open → already recording → speak → stop. Push-side complement to the pull-side plumbing this memo proposes.