Your First Index

What happens when you index a repo

km index ~/my-project

Knowledge Master performs these steps:

1. Discovers files — walks git-tracked files, skips node_modules/.venv/build
2. Parses by type — code files split by function, markdown by heading
3. Embeds chunks — generates 768-dim vectors via Ollama (local)
4. Extracts intelligence — detects tech stack, services, conventions
5. Builds graph — creates nodes and relationships

What gets created in the graph

📦 Repo (my-project)
├── 🔧 USES_TECH → Python, FastAPI, Docker, ...
├── ⚙️ DEFINES_SERVICE → api, postgres, redis
├── 📏 FOLLOWS → snake_case, src/ directory, separate tests
├── 📄 Documents (your files)
│   ├── 🧩 Chunks (with embeddings)
│   └── 👤 AUTHORED_BY → Person (from git history)
└── ⚙️ Service DEPENDS_ON → other services

Supported file types

Extension	Chunking strategy
.py	By function/class definition
.ts, .tsx, .js	By export/function/class
.rs	By fn/impl/struct
.go	By func
.md, .markdown	By heading section
.yaml, .yml	Full file (configs)
.json, .toml	Full file
.txt	By paragraph

Intelligence auto-detection

What	How it's detected
Languages	Presence of pyproject.toml, package.json, Cargo.toml, go.mod
Packages	Parsed from dependency files
Services	docker-compose.yml service definitions, K8s Deployments
Dependencies	depends_on in docker-compose
Conventions	File naming patterns, folder structure, design patterns in code
Authors	Git commit history (last 100 commits)

Tips for best results

• Index at the repo root — ensures git history and docker-compose are found
• Index multiple repos — cross-repo intelligence emerges (shared tech, shared authors)
• Re-index after major changes — km index ~/project is idempotent (updates existing)