PEARL Radix

Documentation

Using the library

Download a file

Every file has a stable numeric URL. Downloads support HTTP range requests, so they resume.

# resume-friendly, keeps the server-side filename
curl -LOC - https://radix.pearlmemory.ca/download/42

# wget
wget -c https://radix.pearlmemory.ca/download/42

# parallel chunks
aria2c -x8 -s8 https://radix.pearlmemory.ca/download/42

Verify what you got

Published checksums appear on each model page and in the API. The download response also carries an X-Checksum-Sha256 header. 

sha256sum model-q4_k_m.gguf
# compare against the value shown on the model page

JSON API

Read-only, no key required, CORS open.

GET /api/models.json?q=&modality=&format=&tag=&sort=&page=&per_page=
GET /api/models/<slug>.json

Mirror an entire model with jq:

curl -s https://radix.pearlmemory.ca/api/models/my-model.json \
  | jq -r '.files[].url' \
  | xargs -n1 -P2 curl -LOC -

Filters

  • q — free text across name, publisher, family and summary
  • modality — text, vision, audio, multimodal, embedding, code, other
  • format — gguf, safetensors, onnx, pytorch, archive
  • tag — tag slug
  • sort — recent, downloads, name, params, size

What gets logged

One row per download: file, timestamp, bytes sent, truncated user agent, and a salted SHA-256 of the IP address. Raw IPs are never written to the database, and there are no cookies on the public side of the site.