A TOML-based catalog of AI provider models. Edit TOML under providers/; a build
step compiles everything into dist/models.json, which the Fleetify app fetches
at runtime to populate its provider model pickers — so new models appear without
shipping an app release.
providers/<provider>/provider.toml # provider metadata (source of truth)
providers/<provider>/models/<id>.toml # one file per model; id = filename
schema/ # JSON Schema for provider + model
scripts/ # build + validate (Bun + TypeScript)
dist/ # GENERATED — do not hand-edit
A model's id is its filename relative to models/. Ids containing / use
subfolders (e.g. nvidia-nim/models/meta/llama-4-maverick-17b-128e-instruct.toml
→ meta/llama-4-maverick-17b-128e-instruct).
bun install
bun run validate # validate all TOML against schema/
bun run build # write dist/models.json (+ index.json + per-provider files)
bun test # unit + e2e testsFetch the generated JSON over a CDN:
https://cdn.jsdelivr.net/gh/fleetifydev/models@main/dist/models.json
Pin a tag for stability: …/fleetifydev/models@v1.0.0/dist/models.json.
Smaller fetches: dist/index.json (provider list) and
dist/providers/<id>.json (one provider).
- Add/edit a
.tomlunderproviders/<provider>/models/.nameis the only required field. Omit optional fields (TOML has nonull). - Run
bun run validate. - Run
bun run buildto regeneratedist/, and commit it alongside your change. Open a PR.
# providers/openrouter/models/anthropic/claude-opus-4-8.toml
[extends]
from = "anthropic/claude-opus-4-8" # <provider>/<model-id>
omit = ["cost"] # drop fields after merge (optional)
[fleetify]
group = "via OpenRouter"dist/models.json carries a schemaVersion. Fields are added additively; the
field is bumped only on a breaking shape change. Consumers ignore unknown fields.
MIT — see LICENSE.