How should I maintain the local Wiki?
Keep Wiki notes focused, use Markdown paths and wikilinks, review extracted facts before writing, preserve source sessions, and rebuild the local index.
How do I separate raw sessions from synthesized pages?
Your LLM Wiki is divided into strict layers to maintain a clean single source of truth.
Organize your knowledge base around concrete entities (projects/, topics/, people/). Use tasks.md as the global checklist and chronological index.
Never dump unformatted, sprawling conversation logs directly into your wiki. Your agent works more reliably with structured, synthesized data.
Why: A wiki is a codebase, not a dump site. Mixing raw conversation data with synthesized conclusions breaks the single source of truth and degrades the agent's context window.
How do I schedule Wiki ingestion?
The LLM Wiki pattern requires compiling knowledge into discrete structures rather than relying on on-the-fly retrieval.
Configure your agent or a local cron job to run scheduled ingestions that read the incoming stream of spoken context and update the corresponding entity pages.
Do not rely on zero-shot RAG (Retrieval-Augmented Generation) against raw transcripts for complex, project-spanning queries.
Why: Raw transcripts contain repetition and conversational noise. Reviewed topic pages reduce the amount of text an agent must load while preserving links to the source sessions.
How do reviewed notes compound?
Questions you ask an agent shouldn't vanish into chat history. Each answer can become a useful synthesis of the source material.
When your agent generates a high-value synthesis (e.g., a technical architecture comparison or a project retrospective), explicitly instruct it to write the output back to the Local Wiki as a new Markdown page.
Why: This creates a "knowledge flywheel." Your analytical explorations become permanent, version-controlled assets rather than ephemeral chat messages.
Which standard Markdown tools can I use?
The LLM Wiki isn't a black box; it's a standard directory. Use the broader ecosystem to enhance your workflow.
Open the local LLM Wiki directory directly in Obsidian. Use native plugins such as Graph View (to visualize dependencies), Dataview (to query YAML front matter), and Web Clipper.
Why: The Local Wiki uses standard Markdown with YAML front matter. Obsidian can read those files directly, while OneMira maintains its own local index for search and backlinks.