Example transcript — this is what your transcript files should look like.

Save each transcript as a plain .txt file inside your main site's `.claude/transcripts/` folder. Use a lowercase, hyphen-separated slug as the filename. For example: `hermes-second-brain.txt`, `openclaw-mission-control.txt`, or `claude-code-seo-agent.txt`.

The filename slug becomes the article URL when Claude writes the post.

The content can be a raw video transcript or a polished version — Claude reads the whole thing and pulls out the key features, numbers, examples, and terminology.

Below is a sample transcript snippet so you can see the format. Yours will be much longer (a typical 10-minute video transcript is 1,500-2,500 words).

─── SAMPLE TRANSCRIPT BELOW ──────────────────────────────────────────────────

Hey everyone, Julian here. Today I want to show you Hermes Second Brain — the new memory system that turns Hermes into a persistent AI assistant that actually remembers everything.

So the problem with most AI agents is they forget. You start a session, you tell them everything about you and your work, and then the next session it's like you've never met. Frustrating.

Hermes Second Brain fixes that. It runs locally on your machine. It indexes every conversation. It builds a vector store. And then when you start a new chat, it pulls relevant memories from your past automatically.

I've been running it for three weeks now. Here's what I've found.

First, the setup is dead simple. You install it via the Hermes CLI with one command. It creates a local SQLite database in your Hermes home folder. No cloud. No accounts. Nothing leaves your machine.

Second, the memory recall is fast. Sub-100ms queries on a database with 1,200+ entries. You don't notice any lag in the chat.

Third — and this is the killer feature — it works across all your Hermes sessions. So I can have a coding session in the morning. A planning session at lunch. A research session in the evening. And the next day, Hermes pulls context from all three.

Here's a concrete example. Last week I was building a new feature. I asked Hermes "what was that library I was using for the kanban board?" Three weeks ago I'd told it I was using DragDropContext. It pulled that memory back instantly. No context dump, no re-explanation needed.

The technical setup: it uses local embeddings via the all-MiniLM-L6-v2 model. 384 dimensions. Stored as quantised int8 to save space. The full memory store for 1,200 entries is about 4MB on disk.

If you want to try it: hermes second-brain init in your terminal. That's it. It starts indexing automatically. Within an hour you'll have a working second brain.

For people in the AI Profit Boardroom, I've put together a setup guide that walks you through every step plus how to integrate it with the Agentic OS dashboard so you can search your second brain from a beautiful UI.

That's it for today. If this video was useful, hit subscribe. And if you want the full setup with everything wired together, AIPB is the place.

─── END SAMPLE ──────────────────────────────────────────────────────────────

Notice the structure:

- Clear opening that sets the problem
- Personal experience and stats ("three weeks", "1,200+ entries", "sub-100ms")
- Concrete example ("last week I was building a feature…")
- Technical details ("all-MiniLM-L6-v2", "384 dimensions", "4MB")
- One-line CTA at the end

When Claude reads this, it picks up:

- The product name → "Hermes Second Brain"
- The benefit → "persistent memory that works across sessions"
- The stats → "1,200 entries, 4MB, sub-100ms, three weeks of use"
- The technical specs → "local SQLite, all-MiniLM-L6-v2 embeddings"
- The setup command → "hermes second-brain init"
- The CTA → AIPB membership

It then uses ALL of these in the 5 unique blog posts it writes.

Your transcripts should aim for the same quality. Concrete. Specific. Numbers. Real examples. Real commands. Real outcomes.

If your transcript is vague, your blog posts will be vague. Garbage in, garbage out.
