Skip to content
Projects

spencernorton/armiger has 97 GitHub stars and 2 forks. GitHot measured 0 stars in 24 hours and 0 in 7 days. Last checked 2026-11-15 16:26 UTC. Last push 2026-11-12.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
0
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

Local ONNX torrent classifier for Bitmagnet and Rats Search, with content type and language prediction and optional metadata enrichment.

Machine Learning

README.md

Original

Armiger

Armiger on NorviTech CI Latest release Hugging Face models Apache-2.0 licence Donate via Stripe

Armiger helps Bitmagnet and Rats Search label torrents by type and language, using names and file lists. Armiger Ultra also adds title matching, metadata-based audio-language estimates and download-availability scores. It runs on a CPU; no GPU or paid API is required.

Published comparison graphics are under provenance review and are not presented here as verified current-release benchmarks. See evaluation status and archived results.

flowchart LR
    Input[Names and file lists] --> Model[Local ONNX classifier]
    Model --> Labels[Type and language predictions]
    Labels --> Ultra[Optional Ultra pipeline]
    Catalog[(Local reference catalogue)] --> Ultra
    Ultra --> Output[Labels and matched metadata]
    Labels --> Output
    Output --> Use[Python API, HTTP or connectors]

Names and metadata go in; labels and optional catalogue matches come out. The classifier runs locally and does not download or inspect media payloads.

Choose your setup

  1. Bitmagnet + Armiger Ultra — set up Bitmagnet + Armiger Ultra, filter via the Torznab proxy (:8097), browse armiger-* tags, and configure all environment variables.
  2. Rats Search + Armiger Ultra — classify items in Rats Search.
  3. Standalone Armiger Ultra — run the ready-made Armiger container on its own.

Choose one guide. The Docker setups include Armiger Ultra, so you do not need to install the standalone version first. Both integration guides start with a fresh, all-in-one installation. Save the Compose file from your guide, choose ARMIGER_UPDATE_CHANNEL=dht, hf or off, and run docker compose up -d; no source build or Python installation is needed.

What Armiger does

Armiger reads names, file paths, sizes and optional swarm counters. It predicts content type, language, subtitle availability and other metadata. Ultra combines those predictions with its catalog and additional checks.

It does not inspect media payloads. Metadata predictions can be wrong and do not prove that a download is safe. See evaluation and limitations for measured results and label coverage. Sharing is off by default.

The model files are 17.6 MB INT8 or 33.3 MB FP32 on disk. These sizes do not describe working RAM: the selected runtime, reference data, batch size and loaded sessions also matter. See resources and measurement scope.

More information

Library 3.12.2 uses model bundle 3.11.0. The Docker image includes verified model files. Python installs download them from a fixed Hugging Face revision and verify them before use. Apache-2.0: license, notices, use policy.


Part of the NorviTech Suite — open-source apps for the Linux desktop and the self-hosted stack

Helios · DH-Waypoint · Armiger · XNote · XNote Placement · SnipSnap · Conductor · NorviOS · Indigo · Road Track · norvitech.com