PLAUD → Obsidian

THE SIMPLE VERSION

Capture automatically. Review context. Then enrich.

The goal is not transcription. The goal is a memory system that gets smarter every time you talk.

Most people record meetings and never use the recordings again. This workflow turns PLAUD transcripts into structured knowledge your AI assistant can actually use.

THE HUMAN WORKFLOW

How the loop works

1. Step 1

   Record with context

   Use PLAUD for calls, meetings, walk-and-talk ideas, conference conversations, and post-call debriefs. Say the who, what, and why out loud at the start.

2. Step 2

   Auto-send the transcript

   Use PLAUD AutoFlow or email export so every recording sends transcript.txt and summary.txt to an inbox your automation can read.

3. Step 3

   Watch the inbox

   Filter for a predictable subject like [Plaud-AutoFlow], check for transcript attachments, and track message IDs so duplicates never slip in.

4. Step 4

   Save raw source

   Store the transcript, summary, and metadata in a stable source folder before any AI rewriting happens.

5. Step 5

   Review context first

   Ask AI to propose who was involved, what the recording was about, what is supported, and what is uncertain before it writes permanent notes.

6. Step 6

   Enrich into memory

   After approval, update meeting notes, people files, project notes, decisions, systems, and content ideas with source-date evidence.

TWO BUILD PATHS

No-code if you want speed. Code if you want control.

The principle stays the same either way: capture automatically, store raw source, review context, then enrich.

FOR YOUR AGENT

Copy this into ChatGPT, Claude, Cursor, or your own AI assistant.

This is the part most people skip: give the agent a clear mission, force it to ask setup questions, and make context review mandatory.

• What email inbox receives PLAUD transcripts?
• What provider is it: Gmail, Outlook, or something else?
• Do you want no-code with Zapier/Make or a coded workflow?
• Where is your Obsidian vault or knowledge base located?
• What folder should raw PLAUD transcripts go into?
• Should permanent notes be created automatically or only after review?
• What subject prefix does PLAUD use? Example: [Plaud-AutoFlow].
• Should the workflow run manually, hourly, daily, or on every new email?

You are reviewing a raw PLAUD transcript before it becomes permanent memory.

Do not write canonical notes yet.
Do not invent names, dates, relationships, decisions, or facts.
Use the transcript as the source of truth.
Use the summary only as orientation.
Separate supported facts from guesses.

Return a context proposal with:
1. Suggested title
2. Recording type: solo voice note, meeting, sales call, client call, interview, unknown
3. Likely attendees and confidence level
4. Related people, projects, companies, and topics
5. Supported facts from the transcript
6. Uncertain facts or possible transcript errors
7. Questions for the human
8. Recommended destination notes or folders
9. Recommended enrichment actions after approval

COMMON MISTAKES

Where these systems go sideways

Transcripts are valuable because they are evidence. The moment the AI starts guessing without review, your knowledge graph starts rotting.

• Treating the AI summary as truth. The transcript is the source.
• Skipping the review gate and letting AI invent context.
• Mixing raw transcripts with permanent notes.
• Forgetting to track processed email IDs.
• Losing the source date when enrichment happens later.

THE RECORDER

Ready to build the loop?

Start with PLAUD, turn on AutoFlow, and make sure your AI assistant reviews context before writing permanent memory.

PLAUD records, transcribes, and summarizes, which gives your AI systems the raw source material they need to build useful memory. The link below is Brian's referral link.