GET /v1/models
GET https://scanner.milgram.dev/v1/modelsNo key is needed. The answer lists the model twice: the release being served, then its alias
nam-shub@latest, with the same terms. The alias’s entry says which release it currently is.
curl https://scanner.milgram.dev/v1/models| Release | nam-shub-v2-20261008 |
| Alias | nam-shub@latest |
| Input | text |
| Output | text: the verdict as JSON, in a chat completion |
| Context length | 500,000 tokens |
Release or alias
Section titled “Release or alias”A request may name either one wherever it names a model: model in a /v1/scan body or query string, or a
chat completion’s model. The alias always means the release being served, so a client that names it keeps
working when a new release replaces the current one; a client that names a release pins it, and is refused
once that release is no longer served. Responses always name the release that answered.
Context length
Section titled “Context length”The context length is the most one request may carry and still be answered without streaming. A non-streaming answer sends nothing until the scan is done, and the networks on the way close a connection that stays silent for about two minutes. Larger inputs, up to the input limit, are scanned when they stream.