Creek-Vault

As of the 2026-07-31 baseline seed (issue #3).

Creek-Vault turns a lifetime of scattered digital exhaust — chat exports, documents, notes, screenshots, messages — into an interlinked Obsidian knowledge base that knows your patterns. It is the desktop, power-user expression of the ecosystem's ontology: every fragment classified by topic, voice, Frequency, archetypal phase, and privacy tier.

What it does for its user

• Ingests almost anything. Eleven source ingestors (Claude/ChatGPT exports, Discord, markdown, PDF/DOCX, spreadsheets, presentations, code, images via OCR, Substack, generic text) plus a read-only Google Drive downloader.
• Protects before it processes. Redaction scans for secrets and PII first; ingestion is consent-gated per source; classification runs locally (Ollama) by default; creek purge delivers right-to-be-forgotten with audit logs.
• Finds the connections. Embedding similarity, temporal proximity, and "eddy" cluster detection interlink fragments across sources.
• Writes back. Index notes, weekly/monthly wavelength reports, a per-frequency Voice Skill Tree, blog-idea mining, and essay drafting in the user's own voice (creek skills / creek mine / creek draft).
• Talks to agents. The creek_mcp server exposes the vault to AI agents, and the crawdad Discord bot is its chat-side interface.

How to use it

Install the CLI, scaffold a vault outside the repo, run the pipeline, and browse the result in Obsidian — Run Creek-Vault locally. The repo's own creek-tools/docs/ guides cover each stage end to end.

Cross-repo relationships

The vault's 05-Wavelength/ and 06-Frequencies/ layers implement the same model as Adepthood's Aspects and the course's stages (docs/Ontology/ is the canonical specification). Adepthood's north star names the Creek Vault MCP seam as part of the app's future: the vault as the deep, local corpus a Higher Self can draw on.