Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
**Major update: Structured multi-table knowledge system and AI-powered routing.** - Adds four new structured tables (`people`, `projects`, `ideas`, `admin`) alongside `thoughts`, supporting relationship, work, and task management. - Implements an AI-driven ingest pipeline: thoughts are embedded, classified with confidence scoring, and routed to a destination table or held in the inbox if confidence is low. - Documents foundational concepts: drop box (inbox), sorter (AI routing), bouncer (confidence threshold), and audit trail for traceability. - Provides cross-table semantic search via a new `search_all` endpoint, as well as per-table semantic search. - Updates references and documentation to detail new schemas, routing rules, and operational building blocks. - Adds conceptual documentation with a new `concepts.md` reference.
What This Skill Does
A personal knowledge system built on Supabase (PostgreSQL + pgvector) and OpenRouter. Captures thoughts from any source, classifies them with an LLM, and routes them into five structured tables: thoughts, people, projects, ideas, and admin. Semantic search is available per table and across all tables via vector embeddings.
Combining pgvector embeddings with structured LLM routing means you can search relationally and semantically across categories without managing separate tools for contacts, tasks, and notes.
When to use it
- Auto-filing a conversation note about a contact under their person record
- Capturing a project idea mid-session and routing it to the ideas table
- Running a semantic search across people, projects, and ideas with a single query
- Tracking pending tasks with due dates through the admin table
- Reviewing the full audit trail of what was captured and where it was filed
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Log that Sarah is thinking about leaving her job to start consulting
- 1Embeds the text via OpenRouter using text-embedding-3-small
- 2Classifies the thought via OpenRouter LLM, extracting type (person_note), topics, people mentioned, and confidence score
- 3Stores the thought in the thoughts table as an audit record
- 4Checks the confidence threshold — if >= 0.6, routes to the people table
- 5Upserts a record for Sarah in the people table with the captured context
Confirmation that the note was logged in thoughts and routed to the people table under Sarah's record
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.