Garth Schwer

Recent Writing

AI-Augmented Digital Gardening

2 min read

AI-Augmented Digital Gardening

Definition: A structured workflow that utilizes Large Language Models (LLMs) to accelerate the maturity of notes from rough ideas to polished, evergreen content.

The Maturity Pipeline

This workflow treats AI not as a writer, but as a specialized editor that helps move notes through specific stages of the Digital Garden maturity model:

  1. Seedling Phase (Capture):
    • Human Action: Rapidly capturing fleeting thoughts, bullet points, or questions.
    • AI Role: None yet. Speed is the priority.
  2. Budding Phase (Expansion):
    • Human Action: Reviewing the seedling and identifying gaps.
    • AI Role: The “Gardener.” Using a specific prompt (or Gemini Gem) to structure the raw bullets, add context, and suggest Wiki-Links to existing concepts.
  3. Evergreen Phase (Polishing):
    • Human Action: Final verification, voice refinement, and publishing.
    • AI Role: The “Copy Editor.” Checking for tonal consistency and clarity without altering the core insight.

Architecture: Subscription vs. API

The implementation of this workflow presents a trade-off between convenience and control.

Option A: The “Gems” Approach (Gemini Advanced)

Google’s Gemini Gems allow for pre-prompted “personas.” You can create a “Garden Architect Gem” pre-loaded with your specific style guide and formatting rules.

  • Pros: Zero technical setup; excellent for the “Budding” phase where creative expansion is needed.
  • Cons: Requires a monthly subscription (~$20/mo); content involves copy-pasting out of your vault; data lives in Google’s ecosystem.

Option B: The RAG Approach (Obsidian + API)

To replicate the experience of NotebookLM—where the AI “knows” your entire library—inside Obsidian, you cannot rely on the standard chat API alone. You need Retrieval-Augmented Generation (RAG).

  • Mechanism: A plugin (e.g., Smart Connections or Text Generator) indexes your vault, finds relevant notes, and feeds them to the API as context.
  • Economics: The Gemini 1.5 Pro API offers a generous free tier for low-volume users. Even the paid tier is “pay-as-you-go,” which is often significantly cheaper than a flat monthly subscription for text-only workflows.

I wonder…

  • Is the friction of copy-pasting into a Gemini Gem worth the superior reasoning capabilities of the 1.5 Pro model compared to lighter models integrated directly into Obsidian?
  • Could I use a local LLM (via Ollama) for the “Seedling” expansion to maintain total privacy, only using the Gemini API for the final “Evergreen” polish?
  • How do I standardise the “Prompts” across different stages so the AI doesn’t hallucinate a new format every time?

References

Graph View