Skip to content

CLI reference

Installing RAGMill puts a ragmill command on your PATH (via the project.scripts entry point). Every command reads its configuration from environment variables or a .env file — see Configuration.

ragmill <command> [args]
Command Description
ragmill ingest <dir> Full ingest + embed + store (no change detection — re-runs duplicate data)
ragmill sync <dir> Incremental sync: add new, update changed, delete removed
ragmill search <query> [--top-k N] Semantic search over stored chunks
ragmill chat [--top-k N] Interactive terminal Q&A over stored chunks
ragmill setup-chat [--yes] Install the local chat model runtime (prebuilt wheel, prompts first)
ragmill count Print the number of stored chunks
ragmill serve [--reload] Start the FastAPI REST server
ragmill export <path.jsonl> Export the whole store to JSONL
ragmill import <path.jsonl> Import JSONL into the store
ragmill configure Launch the standalone setup UI that writes .env
ragmill --version Print the installed RAGMill version and exit

Checking the installed version

ragmill --version          # e.g. "ragmill 0.4.1"

Or from Python:

import ragmill
print(ragmill.__version__)

Common flows

Index and persist a folder

export RAGMILL_SQLITE_PATH=./ragmill.db     # otherwise the store is in-memory
ragmill sync ./my_docs
ragmill count

ragmill search "termination clause" --top-k 5

Chat in the terminal

ragmill chat --top-k 5
# you> summarize the refund policy
# ragmill> Refunds are issued within 14 days of purchase. ...
#          Sources: policy.pdf
# (Ctrl+D or type 'exit' to quit)

Serve the API

ragmill serve                 # binds RAGMILL_HOST:RAGMILL_PORT (default 0.0.0.0:8000)
ragmill serve --reload        # dev mode: hot-reload on code changes

ingest vs sync

  • ingest always processes and inserts everything. Running it twice on the same folder stores every chunk twice.
  • sync tracks per-file hashes and only touches what changed. Prefer sync for anything you re-run. ingest exists for one-shot loads.

Notes

  • search and chat embed your query with the same model used at index time, so they trigger the model download on first use if it isn't cached yet.
  • chat's backend (local / Gemini / OpenAI) is chosen by RAGMILL_CHAT_BACKEND — see Chat & answer generation.