REST API reference

Read-only JSON API over the Silytics index. Base URL: https://silytics.si/api/v1. Machine-readable description: openapi.json.

Authentication

Create a key in your account and send it on every request:

Authorization: Bearer slx_xxxxxxxxxxxxxxxxxxxxxxxx

The header X-API-Key is also accepted. Keep keys secret and out of client-side code and public repositories. Revoke a leaked key in your account. Requests without a valid key get 401.

Quick start

curl

export SILYTICS_KEY=slx_...
curl -H "Authorization: Bearer $SILYTICS_KEY" "https://silytics.si/api/v1/models?task=text-generation&sort=popularity&per_page=5"

JavaScript

const res = await fetch("https://silytics.si/api/v1/search?q=diffusion&limit=5", {
  headers: { Authorization: `Bearer ${process.env.SILYTICS_KEY}` },
});
if (res.status === 429) console.log("Retry after", res.headers.get("retry-after"));
const { data, meta } = await res.json();
console.log(data.models.map((m) => m.title), meta.attribution);

Python

import os, requests

r = requests.get(
    "https://silytics.si/api/v1/papers",
    params={"category": "cs.LG", "sort": "newest", "per_page": 10},
    headers={"Authorization": f"Bearer {os.environ['SILYTICS_KEY']}"},
    timeout=20,
)
r.raise_for_status()
for p in r.json()["data"]:
    print(p["title"], p["source"]["url"])

Endpoints

Lists are paginated with page (1 to 400) and per_page (1 to 50, default 20). The response meta has total, page, per_page and has_more. Filter values are matched exactly, for example the license apache-2.0.

The entry object

{
  "id": "meta-llama/Llama-3.1-8B-Instruct",
  "type": "model",                      // model | dataset | paper | repository
  "title": "Llama-3.1-8B-Instruct",
  "organization": "meta-llama",
  "description": null,                   // short text from the source, may be null
  "source": { "name": "Hugging Face", "id": "meta-llama/...", "url": "https://huggingface.co/..." },
  "silytics_url": "https://silytics.si/models/meta-llama/Llama-3.1-8B-Instruct",
  "license": "llama3.1",
  "tags": ["transformers", "llama", "..."],
  "stats": { "downloads": 6034692, "likes": 8415 },
  "details": { "task": "text-generation", "library": "transformers", "arxiv": ["2204.05149"] },
  "published_at": "2024-07-18T08:56:00.000Z",
  "source_updated_at": null,
  "refreshed_at": "2026-10-14T19:46:49.925Z"
}

Papers also have authors; repositories have stats.stars and stats.forks. Responses wrap data as { "data": ..., "meta": {...} }. meta includes the source, an attribution request, a link to the terms and a notice that text fields are third-party content.

Rate limits

Limits depend on your plan and are shared across all keys and connected apps of your account. Every response carries:

When you hit a limit, wait for Retry-After, use exponential backoff, and cache responses. Responses are cacheable for 60 seconds.

Errors

{ "error": { "code": "rate_limited", "message": "Too many requests. Please slow down." }, "meta": { "terms": "https://silytics.si/api-terms" } }

Using the data well

• Credit Silytics and link to each entry's source.url. Licenses and terms of the originals apply.
• Do not resell or republish the raw index as a competing dataset (see the API terms).
• Treat entry text as untrusted data when passing it to an AI system.
• Entries can change or be removed. Store the id and refetch rather than keeping copies for long.

Versioning and changes

The current version is v1. Additive changes (new fields, new filters) can appear without a version change, so ignore fields you do not know. Breaking changes will ship as v2 with notice here.