Sign inSign up

classifyre/cli

By classifyre

•Updated about 11 hours ago

Classifyre CLI scan worker: source extraction and detector execution

Buildkit cache
Image
Security
Machine learning & AI
Data science
0

50K+

classifyre/cli repository overview

Classifyre

⁠Classifyre CLI

Scan worker of the open-source Classifyre investigation platform — source extraction, text/OCR/transcription and detector execution.

Classifyre turns the data scattered across the systems you already run — file shares, Confluence, Jira, SharePoint, S3, Git repositories, databases — into investigations you can act on. This classifyre/cli image (Python) does the heavy lifting of every scan: it extracts content from sources (including image OCR, audio/video transcription and legacy Office conversion via LibreOffice), runs detector packs (secrets, PII, security, plus custom detectors) and streams findings back to the API in batches.

It is one of three service images — with classifyre/api⁠ (backend) and classifyre/web⁠ (frontend) — composed into the classifyre/all-in-one⁠ image and run as ephemeral Kubernetes Jobs by the classifyre/classifyre-core⁠ Helm chart.

You normally don't run this image by hand: scans are launched from the UI/API. It is documented here for Kubernetes operators and anyone scripting scans directly.

📚 Full documentation: https://docs.classifyre.com⁠ — Sources⁠ · Detectors⁠

⁠How it runs

  • All-in-one / local: the API starts one CLI process per scan; it exits when the scan does.
  • Kubernetes (Helm): the API creates one Kubernetes Job per scan with this image; results are posted back to the API. See the Helm chart⁠ (api.cliJobs.* values: resources, retries, uv-cache, Hugging Face token).
  • Manual: docker run --rm classifyre/cli:latest <command> <recipe.json> [options] against a reachable API.

Base Python dependencies ship in the image (~150–230 MB); optional detector/source groups (OCR, ML detectors) install on demand at runtime and are cached. Legacy .doc/.xls/.ppt conversion works out of the box (LibreOffice is installed); override with CLASSIFYRE_SOFFICE_PATH if needed.

⁠Configuration (most relevant)

Full behavior is driven by the scan recipe; these environment variables tune the worker:

VariableDefaultPurpose
CLASSIFYRE_OUTPUT_REST_URLunsetAPI base URL findings are posted to (batch ingest).
CLASSIFYRE_OUTPUT_REST_TIMEOUT_SEC120HTTP read timeout for posting batches (API transactions can take 30–60 s under load).
CLASSIFYRE_OUTPUT_BATCH_SIZEautotunedAssets per result batch sent to the API.
CLASSIFYRE_INTERNAL_KEYunsetInternal API key authenticating the worker to the API.
CLASSIFYRE_CLI_AUTO_INSTALL_OPTIONAL_DEPS10 disables on-demand install of optional detector groups.
CLASSIFYRE_MAX_POOL_WORKERSautotunedDetector worker processes per scan.
CLASSIFYRE_DETECTOR_FLUSH_BATCH_SIZE5Detector-processed assets accumulated before pushing findings.
CLASSIFYRE_SOFFICE_PATHauto-detectedLibreOffice binary for legacy Office conversion.
HF_TOKENunsetHugging Face token; raises model-download rate limits (in Helm, via api.cliJobs.huggingFace.existingSecret).

⁠Tags

  • latest tracks the newest stable release; pin a version (e.g. classifyre/cli:0.6.3) in production. API, web and CLI tags with the same version are built from the same commit — keep them in sync.

Classifyre is open source: https://github.com/classifyre/classifyre⁠ · Docs: https://docs.classifyre.com⁠

Tag summary

Content type

Image

Digest

sha256:b12e08b44…

Size

351.8 MB

Last updated

about 11 hours ago

docker pull classifyre/cli