Every public record is a URI, and ant dereferences it.

ant is a single pure-Go binary that puts one address space over the whole tamnd/*-cli family. A book on Goodreads, an account on X, an article on Wikipedia, a video on YouTube — each is a short URI like goodreads://book/2767052, x://user/nasa, wikipedia://page/Alan_Turing, or youtube://video/dQw4w9WgXcQ. Hand one to ant and it fetches the record, follows the typed links out of it, and writes the slice of the graph you asked for to disk, no matter which site each name lives on.

No API key, no daemon, nothing to run alongside it. The site CLIs are linked in as Go libraries, so ant is one static binary that already knows how to read every site it was built with.

ant get goodreads://book/2767052          # fetch and print the record
ant resolve "https://x.com/nasa"          # x://user/nasa
ant links goodreads://book/2767052        # the outbound edges, as URIs
ant export goodreads://author/153394 --follow 1   # write the subgraph to disk

go install github.com/tamnd/ant/cmd/ant@latest

Or grab a prebuilt binary from the releases, or run the container image:

docker run --rm ghcr.io/tamnd/ant:latest --help

A resource URI is scheme://authority/path. The scheme picks the site, the authority picks the record type, the path is the id:

goodreads://book/2767052
goodreads://author/153394
x://user/nasa
x://status/20
wikipedia://page/Alan_Turing
wikipedia://category/Computability_theory
youtube://video/dQw4w9WgXcQ
youtube://channel/UCuAXFkgsw1L7xaCfnd5JJOw

Every verb takes a URI and does one thing:

And the graph verbs work over the on-disk URI tree under $HOME/data, where a record's file path is its URI:

Resolve a messy URL to its URI, then dereference it:

$ ant resolve "https://www.goodreads.com/book/show/2767052-the-hunger-games"
goodreads://book/2767052

$ ant get goodreads://book/2767052
{
  "@id": "goodreads://book/2767052",
  "@type": "goodreads/book",
  "@fetched": "2026-06-14T09:31:00Z",
  "@links": {
    "author": ["goodreads://author/153394"]
  },
  "data": { ... }
}

Cross a site boundary without changing tools — the same resolve/get work on X:

$ ant resolve 20 --on x
x://status/20
$ ant url x://user/nasa
https://x.com/nasa

Walk the graph and write it to disk as the URI tree:

$ ant export goodreads://author/153394 --follow 1 --to ./data
$ ant ll goodreads:// --data ./data
goodreads://author/153394
goodreads://book/2767052
...
$ ant graph goodreads://author/153394 --depth 1 --format dot | dot -Tsvg > graph.svg

Serve the namespace as dereferenceable linked data:

$ ant serve --addr :7777 &
$ curl -s localhost:7777/goodreads://book/2767052 | jq .['@type']
"goodreads/book"
$ curl -s 'localhost:7777/resolve?input=https://x.com/nasa'
{"uri":"x://user/nasa"}

ant learns a new site the way database/sql learns a new database — a blank import that registers a driver. The site CLI exposes a Domain, and ant links it in with one line in cli/root.go:

import (
    _ "github.com/tamnd/goodread-cli/goodread"
    _ "github.com/tamnd/wikipedia-cli/wiki"
    _ "github.com/tamnd/x-cli/x"
    _ "github.com/tamnd/ytb-cli/youtube"
)

The driver declares its records with struct tags — kit:"id" marks the identifier, kit:"link,kind=goodreads/author" marks an edge, kit:"body" marks the prose — and the verbs above work for free. See 8000_uri_drivers.md for the mechanism.

cmd/ant/   thin main, wires cli.Root into fang
cli/                 the cobra command tree, one file per verb group
ant/                 the library: the Engine over a kit.Host of registered domains
docs/                tago documentation site

make build      # ./bin/ant
make test       # go test ./...
make vet        # go vet ./...

Push a version tag and GitHub Actions runs GoReleaser, which builds the archives, Linux packages, the multi-arch GHCR image, checksums, SBOMs, and a cosign signature:

git tag v0.1.0
git push --tags

The Homebrew and Scoop steps self-disable until their tokens exist, so the first release works with no extra secrets.

Apache-2.0. See LICENSE.