Appearance
Support
This page is read from the declarations the compiler enforces, not maintained by hand: the provider set is SERVED_PROVIDERS (what the normalize gate accepts) and the cloud matrix is CLOUD_TARGETS (what the infra planner refuses outside of). If a provider or cloud is missing here, a spec that names it is rejected at plan time rather than quietly generated.
Model providers
Every provider below is served by py-fastapi and ts-vercel and is accepted by the normalize gate.
| Provider | Served by |
|---|---|
anthropic | py-fastapi, ts-vercel |
bedrock | py-fastapi, ts-vercel |
gemini | py-fastapi, ts-vercel |
google | py-fastapi, ts-vercel |
openai | py-fastapi, ts-vercel |
openai-compatible | py-fastapi, ts-vercel |
fake | py-fastapi, ts-vercel |
Clouds and deployment targets
A yes means the infra planner emits a module for that target on that cloud. A dash means it refuses the plan with a named diagnostic; it does not emit unvalidated HCL. Every emitted module is byte-asserted against a committed fixture and validated with tofu validate in CI.
| Target | aws | gcp |
|---|---|---|
batch | yes | yes |
container | yes | yes |
edge | yes | — |
serverless | yes | yes |
The emitted infrastructure is OpenTofu. A plan targets one cloud at a time, because two providers in one state is a blast radius rather than a feature.
Spec language
The current spec language version is 0.11. A construct that needs a later version is reported as spec/not-supported, naming the version it needs, instead of being ignored.
Conformance
The corpus holds 77 cases over 41 operations, run against every target. A case that a target cannot satisfy is a declared capability boundary, not a skipped test.
Next: the generator targets, the CLI reference, or the conformance index.