ctx catalog

ctx catalog indexes documentation, generates embeddings via an OpenAI-compatible endpoint, and answers natural-language queries. 100% local.

Usage

# Register a documentation collection
ctx add my-project --source ./docs --include "**/*.md" --exclude "**/node_modules/**"

# Index documents + generate embeddings (requires Ollama)
ctx index my-project --with-embed

# Search by intent
ctx search my-project "how does the auth pipeline work?"

Subcommands

Command	Function
ctx add <name>	Register a new collection (sources, include/exclude patterns, models)
ctx index <name>	Index documents (detects new/modified)
ctx embed <name>	Generate embeddings for pending chunks
ctx search <name> <query>	Semantic + lexical search
ctx list	List all collections
ctx status <name>	Show collection stats
ctx compact <name>	Optimize storage
ctx bootstrap --path <dir>	One-shot: add + index

Search modes

The query prefix selects the mode:

ctx search my-project "exact: middleware chain"          # lexical only
ctx search my-project "conceptual: auth flow"            # semantic emphasis
ctx search my-project "expanded: how do hooks work?"     # query expansion (planned)
ctx search my-project "natural query"                    # auto-detect (default)

Configuration

Global (~/.ctx/config.toml):

[llm]
endpoint = "http://localhost:11434"
embedder = "nomic-embed-text"
reranker = "llama3.2"

Per-collection override (set during ctx add):

ctx add my-project \
  --source ./docs \
  --embedder-model bge-base-en \
  --reranker-model llama3.2 \
  --llm-endpoint http://localhost:11434

Without Ollama

ctx search falls back to lexical (BM25) mode automatically. Useful for environments where you can’t run a local LLM but still want semantic-ish search via term matching.