name: kindex
audience: public
description: >
  Persistent knowledge graph that remembers across AI sessions.
  Hybrid FTS5+graph search, MCP plugin for Claude Code, CLI (kin),
  29 MCP tools. SQLite-backed with vector search. The memory layer
  for all projects.

domains:
  - knowledge-graph
  - ai-tooling
  - mcp
  - context-management
  - search
  - python

voice:
  tone: precise, tool-oriented, developer-first
  principles:
    - Search before add — no duplicates
    - Graph value is in connections, not isolated nodes
    - Watches and reminders surface proactively
    - Modes scope context to current work
    - Decay keeps the graph fresh — unused nodes fade
  anti-patterns:
    - Adding nodes without linking them
    - Storing ephemeral task state as permanent knowledge
    - Ignoring decay — stale nodes pollute context

standards:
  code_review:
    - 930+ tests, all must pass
    - Pydantic models for all data structures
    - SQLite migrations are forward-only
    - MCP tools match CLI capabilities
  documentation:
    - Docs site at jmcentire.github.io/kindex/

connects_to:
  - pact
  - signet
  - stigmergy
  - chronicler
  - apprentice
  - sentinel

work_policy:
  require_active_tag: true
  linear:
    enabled: false
    require_issue: false
  git:
    block_commit_without_tag: false
    block_commit_without_linear: false
    block_push_without_tag: false
    block_push_without_linear: false
