Chromy/README.md

Chromy

Chromy is small and simple to use command-line utility for working with a local Chroma database. It lets you create collections, ingest files as chunked embeddings, and run similarity queries against stored documents. It integrates perfectly with agentic coding tools via simple skills (see an example in the skills directory).

What it does

• manages local Chroma collections
• chunks files with semchunk
• generates embeddings with Chroma's default embedding function
• stores chunk text plus source file metadata
• queries collections and prints readable results

Requirements

• Python 3.12+
• a local environment able to install the project dependencies in pyproject.toml

Libraries

Runtime libraries

• chromadb — persistent vector database used to store collections, embeddings, documents, and metadata.
• openai — dependency used by the embedding stack for model and API integrations.
• pymupdf4llm — extracts text from PDF documents for ingestion.
• python-dotenv — loads environment variables from local .env files.
• rich — provides styled terminal output and progress bars.
• semchunk — splits source documents into chunks before embedding.
• tiktoken — tokenization support used during chunking and embedding preparation.
• transformers — model and tokenizer support used by the embedding pipeline.
• typer — powers the CLI commands and argument parsing.

Development libraries

• mypy — static type checking.
• nuitka[onefile] — builds standalone one-file executables.
• pytest — test runner for the project.
• ruff — linting and formatting.

Installation

For local development, install the project dependencies with uv:

uv sync

Or with pip:

python -m venv .venv
source .venv/bin/activate
pip install -e .

Build

Build the source distribution and wheel with uv:

uv build

The build artifacts are written to dist/.

Install as a tool with uv

The project exposes a chromy command through the Python packaging entrypoint. Install it as a standalone uv tool from the project directory:

uv tool install .

After installation, run the CLI directly:

chromy --help

To install from a built wheel instead:

uv build
uv tool install dist/chromy-1.0.0-py3-none-any.whl

During development, install the tool in editable mode so changes in the working tree are picked up without reinstalling:

uv tool install --editable .

Running the CLI

The project entrypoint is available as the chromy command after installing the tool:

chromy --help

You can also run it from the source tree without installing the tool:

uv run python -m chromy.main --help

Chroma storage location

By default, Chromy uses Chroma's default persistent location behavior (a local chroma/ directory based on your current working directory when you run the command).

You can override this with CHROMA_FOLDER.

• CHROMA_FOLDER must point to a parent directory.
• Chromy will store data in <CHROMA_FOLDER>/chroma.
• Relative paths are supported and are resolved from the current working directory.
• If CHROMA_FOLDER is set, it takes precedence over the default behavior.
• If the configured location is invalid or not writable, the command fails with an explicit error (no fallback to the default location).

Setting the variable once in .zprofile or .profile ensures a consistent usage of the variable.

Examples:

# absolute parent path
CHROMA_FOLDER=/tmp/chromy-data chromy list-collections

# relative parent path (resolved from current directory)
CHROMA_FOLDER=.local-data chromy create-collection notes

Running Tests

Run the test suite with pytest:

uv run pytest -q

Development Checks

Run Ruff linting:

uv run ruff check .

Check Ruff formatting:

uv run ruff format --check .

Run static type checking with mypy:

uv run mypy .

Commands

list-collections | lc
create-collection <collection> | cc <collection>
delete-collection <collection> | dc <collection>
count <collection> | c <collection>
import <collection> <file> [<file> ...] | i <collection> <file> [<file> ...]
query <collection> <query_text> | q <collection> <query_text>
delete <collection> --where <condition>=<value> | del <collection> --where <condition>=<value>

Aliases

• lc → list-collections
• cc → create-collection
• dc → delete-collection
• c → count
• i → import
• q → query
• del → delete

Examples

Create a collection:

chromy create-collection notes
# alias
chromy cc notes

Add one or more files:

chromy import notes ./docs/example.txt
chromy import notes ./docs/intro.md ./docs/setup.md
chromy import notes *.md
# alias
chromy i notes ./docs/example.txt

Import a large batch of files with find:

find ./docs -type f \( -name '*.md' -o -name '*.txt' \) -exec chromy import notes {} +

Count stored records:

chromy count notes
# alias
chromy c notes

Search the collection:

chromy query notes "How do I configure this project?"
# alias
chromy q notes "How do I configure this project?"

List collections:

chromy list-collections
# alias
chromy lc

Delete a collection:

chromy delete-collection notes
# alias
chromy dc notes

Delete records by metadata:

chromy delete notes --where file_name=example.txt
# alias
chromy del notes --where file_name=example.txt

How ingestion works

When you run import, each file is:

1. read from disk
2. split into chunks
3. embedded
4. inserted into the target collection with the original file path stored as metadata

Query results include the stored document chunk, its id, distance, and file name when available.

Notes

• by default, collections are stored in a local persistent Chroma database in the current directory
• set CHROMA_FOLDER to override the parent location; Chromy will use <CHROMA_FOLDER>/chroma
• import requires the target collection to already exist
• import accepts one or more file paths
• unquoted glob patterns such as *.md are expanded by the shell before chromy starts
• quoted glob patterns such as "*.md" are treated as literal paths and are not expanded by chromy
• unmatched unquoted globs may behave differently by shell: zsh commonly fails before chromy starts, while bash may pass the literal pattern through depending on shell settings
• the CLI reports file-specific import failures and continues with the remaining files
• when importing multiple files in an interactive terminal, the CLI shows a Rich progress bar