Hosted API¶

Switched off

The hosted API is disabled. It runs on a hosting plan that allows less compute per request than the engine needs to start, so about half of all requests failed, and a coin flip is worse than a plain refusal. Every route now answers with a message that says so, and the badge answers the gray n/a. The reference below describes what the API does when it is on. Everything else runs the engine on your own machine and is unaffected: the web app, the command line tool, the editor integrations, the Obsidian plugin, the MCP server, and the slop-chop-wasm package for Node.

Chop text over HTTP. The API runs the same deterministic rules engine as everything else, compiled to WebAssembly on Cloudflare Workers. No model, no account, no storage: the text is processed in memory and the response is the only thing that leaves. Same input, same output, every time.

Base URL: https://api.slop-chop.com

POST /chop¶

curl -s https://api.slop-chop.com/chop \
  -H 'Content-Type: application/json' \
  -d '{"text": "In summary, we leverage a myriad of robust tools."}'

{
  "output": "We use many solid tools.",
  "findings": [ { "rule": "phrase:in summary,", "match": "In summary, w", "offset": 0 } ],
  "remaining": [],
  "score":      { "value": 80, "scoringVersion": 1 },
  "scoreAfter": { "value": 0, "scoringVersion": 1 }
}

A finding with a replacement was rewritten and one without was only flagged. remaining is what a check reports on the output, the tells no rule can rewrite, so a score that stays high after a chop comes with the list of what is left. Compare scores only within one scoringVersion, see Scoring versions.

The body takes the same options as the npm package:

Field	What it does
`text`	The text to chop. Required, up to 1MB.
`presets`	Built-in preset names to apply. Defaults to `["cleaver"]`.
`voice`	`{keep, prefer, avoid}` folded on top, your swaps winning.
`profile`	A full profile that replaces the built-in default.

With a voice:

curl -s https://api.slop-chop.com/chop \
  -H 'Content-Type: application/json' \
  -d '{"text": "we leverage robust tools",
       "voice": {"keep": ["robust"], "prefer": {"leverage": "wield"}}}'

{ "output": "we wield robust tools" }

GET /presets¶

Lists the built-in preset names.

GET /badge¶

Scores a public repository's README and answers with an SVG badge, for embedding in that README.

[![slop score](https://api.slop-chop.com/badge?repo=dcadolph/slop-chop)](https://slop-chop.com)

Field	What it does
`repo`	The repository as `owner/name`. Required. Public repositories only.

The README is read from GitHub's raw host, scored with the default profile, and discarded. Only the first 16KB is scored, to keep the request inside the compute a hosted Worker is allowed. Nothing is stored. The label is fixed, so the endpoint renders a score and nothing else.

A repository that cannot be read, has no README, or is private answers the gray n/a badge with a 200, because a badge is loaded as an image and a non-200 renders as a broken image. Scored badges are cacheable for six hours and the gray one for five minutes, which GitHub's image proxy honors.

What the number measures is worth being plain about: READMEs score low. Across a sample of widely used repositories the scores landed between 0 and 27 out of 100, because the score is a density over prose and a README is mostly headings, lists, links, and fenced code, which the engine skips. Treat the badge as a published commitment rather than a discriminating test, and use slop-chop check in CI on the prose files where the measurement has room to move.

Notes¶

CORS is open, so a browser page can call it directly.
A body over 1MB answers 413. Malformed JSON answers 400.
The optional model rewrite is not part of the API. It stays where your keys stay: the CLI and the web app. For private text, prefer those. They never send text anywhere at all.