CLI Application

Python CLI for source extraction, detector execution, and batched output delivery.

Setup

cd /unstructured/apps/cli
uv sync
# Optional if you want an activated shell instead of `uv run ...`:
source .venv/bin/activate

Optional detector groups:

uv sync --group detectors
# or specific groups: --group secrets --group pii --group threat ...

System dependency: LibreOffice

Legacy Office formats (.doc, .xls, .ppt) have no usable pure-Python parser, so src/utils/legacy_office.py shells out to soffice --headless --convert-to docx|xlsx|pptx and feeds the result through the normal OOXML extractors. Without it those files scan as "no content available" — a silent coverage gap, not an error.

Kubernetes: installed in the CLI image (libreoffice-*-nogui), so this works out of the box.
Desktop and local dev: install it yourself — brew install --cask libreoffice (macOS), apt install libreoffice-writer libreoffice-calc libreoffice-impress (Debian/Ubuntu), or libreoffice.org on Windows. The desktop app deliberately does not bundle LibreOffice: it is ~550 MB even stripped, and slimming the upstream bundle breaks its code signature, which Apple Silicon punishes by killing the process on launch.
Non-standard install: set CLASSIFYRE_SOFFICE_PATH to the binary. The CLI checks that first, then PATH, then the standard macOS, Windows, and Linux-tarball install locations.

When no binary is found the file is not silently reported as empty. The scan raises an ENGINE_UNAVAILABLE text-extraction coverage error that fails that asset's text scan, lands in scan_stats.errors and the run's text-coverage counters, and carries a platform-specific message telling the user how to install LibreOffice.

tests/utils/test_legacy_office.py skips its round-trip tests when no LibreOffice is found, so a green local run does not by itself prove this path works.

Command Syntax

Use the thin wrapper:

uv run main.py <command> <recipe.json> [options]

Or direct module entrypoint:

uv run python -m src.main <command> <recipe.json> [options]

Commands:

test - test source connection.
discover - discover source resources.
extract - run extraction and emit batched output.
evaluate-file - internally evaluate detectors against one local file.

Extract Output Model

Extraction always emits in batches. Recipes do not contain output configuration; output is controlled by CLI flags and environment variables.

Output types:

console - emits NDJSON envelopes to stdout.
file - appends NDJSON envelopes to a file.
rest - pushes batches to API endpoints and finalizes run.

Default behavior:

If source_id is present (--source-id or SOURCE_ID env), default output is rest.
Otherwise default output is console.
Default batch size is 20.

CLI Options

Global/common:

--debug - enable debug logging.
--detectors-file <path> - file evaluation only.

Extract output options:

--output-type rest|file|console
--output-batch-size <int>
--output-rest-url <url> (complete API base, including the namespace path)
--output-file-path <path>
--source-id <uuid>
--runner-id <uuid>
--managed-runner (REST only; runner lifecycle managed by API orchestrator)

Environment fallbacks:

SOURCE_ID, RUNNER_ID
CLASSIFYRE_OUTPUT_TYPE, CLASSIFYRE_OUTPUT_BATCH_SIZE
CLASSIFYRE_OUTPUT_REST_URL, CLASSIFYRE_OUTPUT_REST_TIMEOUT_SEC
CLASSIFYRE_OUTPUT_FILE_PATH
API_URL (fallback base URL for REST output)

Practical Examples

1) Console output (quick local test)

uv run main.py extract ./wordpress-recipe.json --output-type console --output-batch-size 1

You will see NDJSON lines like:

{"event":"batch", ...}
{"event":"finish", ...}

2) File output

uv run main.py extract ./wordpress-recipe.json \
  --output-type file \
  --output-file-path /tmp/classifyre-assets.ndjson \
  --output-batch-size 20

3) REST output (manual CLI to backend)

uv run main.py extract ./wordpress-recipe.json \
  --output-type rest \
  --source-id <source_uuid>

Notes:

--runner-id optional for manual runs. If omitted, CLI creates external runner automatically.
--output-rest-url is optional. If omitted, CLI uses CLASSIFYRE_OUTPUT_REST_URL, then API_URL, then http://localhost:8000. In a namespaced deployment, pass the complete base (for example https://api.example/ns-id); the CLI appends all asset, finding, runner, and source endpoints beneath it.
--managed-runner should be used only for API-orchestrated runs where runner already exists.

4) REST output with explicit runner (managed/orchestrated style)

uv run main.py extract ./wordpress-recipe.json \
  --output-type rest \
  --source-id <source_uuid> \
  --runner-id <runner_uuid> \
  --managed-runner

5) Full extract command with all output flags

uv run main.py extract ./wordpress-recipe.json \
  --output-type rest \
  --output-batch-size 20 \
  --output-rest-url http://localhost:8000 \
  --output-file-path /tmp/classifyre-assets.ndjson \
  --source-id <source_uuid> \
  --runner-id <runner_uuid> \
  --managed-runner

Use --output-file-path only when --output-type file.

Dev Scripts

bun run dev - run CLI quickly.
bun run lint - ruff format/check.
bun run check-types - mypy.
bun run test - pytest suite.

Release files for classifyre-cli 0.4.103

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for classifyre-cli 0.4.103
File	Size	Uploaded
classifyre_cli-0.4.103.tar.gz	1.1 MB	Aug 9, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for classifyre-cli 0.4.103
File	Interpreter	ABI	Platform	Reset
classifyre_cli-0.4.103-py3-none-any.whl 515.1 kB Aug 9, 2026	Python 3	none	any	Details

Total release size: 1.6 MB

Release files / classifyre_cli-0.4.103.tar.gz

Download URL	classifyre_cli-0.4.103.tar.gz
Size	1.1 MB
Tags	Source
SHA-256 checksum How to use checksums	`5a320eaf2074f5d36e76e6bb41c7527a32eb45af6a619cf90e94620ca82fa56b`
BLAKE2b-256 checksum How to use checksums	`e04a33bdedd9e6774db07905460fe53d7bfc68247efbe384b609dc009f07a50c`
Upload date	Aug 9, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`uv/0.12.3 {"installer":{"name":"uv","version":"0.12.3","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`

Release files / classifyre_cli-0.4.103-py3-none-any.whl

Download URL	classifyre_cli-0.4.103-py3-none-any.whl
Size	515.1 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`3d52b98e642f02b871c3af23d52396a4f3aa4dd6d54cab51f0316664c79fbeb7`
BLAKE2b-256 checksum How to use checksums	`aea4dbadc046f8d8e62aff1ff893a1f5c72c0e003af427d7f7b5e6b1aacffe17`
Upload date	Aug 9, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`uv/0.12.3 {"installer":{"name":"uv","version":"0.12.3","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`