Use when ingesting sources into a knowledge wiki, maintaining a compounding knowledge base, querying accumulated knowledge, or running wiki health checks. Also use when setting up the LLM wiki stack (obsidian-headless, ByteRover, obsidian-mind) on a new machine.
A compounding personal knowledge base maintained by an AI assistant. The wiki gets richer with every ingest — not a retrieval index, a compounding artifact.
Key distinction: RAG retrieves from scratch every time. This wiki is built once and kept current.
When adding a new source (URL, article, document, market signal):
wiki/sources/<title>.mdwiki/entities/ (create if missing)wiki/index.md — add/modify entrieswiki/log.md — ## [YYYY-MM-DD] ingest | Source TitleA single source typically touches 5-15 pages. This is expected and correct.
Never just create one file and stop. Every ingest must update related existing pages.
wiki/index.md to find relevant pages(source: [[Source Title]])wiki/topics/<topic>.md — don't let synthesis disappear into chat historywiki/log.md — ## [YYYY-MM-DD] query | Question askedRun periodically or when wiki feels stale:
Append results to wiki/log.md.
# [Entity Name]
**Type:** person | company | product | concept
**Last updated:** YYYY-MM-DD
## Summary
One paragraph.
## Key Facts
- Fact (source: [[Source Title]])
## Relationships
- [[Related Entity]] — how they relate
## Timeline
- YYYY-MM-DD: Event (source: [[Source Title]])
## Open Questions
- Question needing a source to answer
## Contradictions
- Claim A (source X) vs Claim B (source Y) — unresolved
# [Source Title]
**Type:** article | paper | announcement | analyst-piece
**URL:**
**Date:** YYYY-MM-DD
## Summary
2-3 sentences.
## Key Takeaways
- Takeaway 1
## Pages Updated
- [[Page]] — what changed
ByteRover handles structured domain/project knowledge (context tree with recency decay).
# Add knowledge
brv curate "factual statement about a project or domain"
# Query knowledge
brv query "what do we know about X?"
# Sync to cloud
brv push
# Check status
brv status
When to use ByteRover vs wiki:
# 1. Install obsidian-headless
npm install -g obsidian-headless
ob login
# 2. Set up vault sync
ob sync-setup --vault "VaultName" --path ~/Vault --device-name "$(hostname)"
# 3. Install sync daemon
cp systemd/obsidian-sync.service ~/.config/systemd/user/
# Edit ExecStart paths to match your node binary location
systemctl --user daemon-reload && systemctl --user enable --now obsidian-sync.service
# 4. Install ByteRover
npm install -g byterover-cli
brv providers connect openai --api-key YOUR_KEY
# 5. Install obsidian-mind (personal memory)
git clone https://github.com/breferrari/obsidian-mind.git ~/Vault-mind
Find the node binary path with: which node
Find the obsidian-headless cli.js path with: npm root -g
The vault's WIKI-SCHEMA.md (or named after your assistant) governs domain-specific update triggers — which pages to touch when specific events happen (competitor funding, benchmark run, positioning decision). Read it at the start of every wiki session.
Repo: github.com/joshuaboys/llm-wiki-stack