Skip to content

Let the live ITs follow TYPESAFE_BASE_URL and document using Laya - #9

Merged
tzolov merged 2 commits into
mainfrom
laya-compat
Sep 29, 2026
Merged

tzolov merged 2 commits into
mainfrom
laya-compat

Conversation

@tzolov

@tzolov tzolov commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

Laya's laya-serve (laya#31) exposes Jev's POST /v1/systemone protocol. This PR makes the project usable against it and documents what to expect.

  • Live ITs follow TYPESAFE_BASE_URL. The ITs pinned .baseUrl(TypeSafeConstants.DEFAULT_BASE_URL), so they could only hit the hosted API. With the pin dropped, TypeSafeClient.builder() falls back to TYPESAFE_BASE_URL, so the same ITs run against any compatible server. The TYPESAFE_BASE_URL warnings in the README and on the starter page now say so.
  • New docs page, Client → Using Laya:
    • running laya-serve, on port 8002 to avoid clashing with mkdocs serve
    • pointing the plain client and the Boot starter at it
    • the protocol differences measured against Laya 0.3.21: no /v1/models, no request id, output_tokens always 0, nouls need instructions, {"detail"} errors without a type, and different validation status codes (a malformed question is 422 where Jev answers 400; a null state is 400 where Jev answers 422)
    • how Laya's answers compare with Jev's on the demos, including its sensitivity to the order of the state's fields
    • criteria that need world knowledge: Laya can't reliably judge physical plausibility, even with the bounds spelled out, so the Model-as-a-judge demo is not a meaningful Laya test; use a code criterion for such checks

Against Laya, the SDK handled every response without changes; all failing ITs are protocol gaps or answer-quality differences, not SDK bugs.

The ITs pinned the hosted base URL, so they could only run against Jev.
Dropping the pin lets TypeSafeClient.builder() fall back to
TYPESAFE_BASE_URL, which makes the ITs runnable against any server
speaking the System One protocol, such as a local laya-serve.

Add a "Using Laya" page under Client: running laya-serve, pointing the
plain client and the Boot starter at it, the protocol differences
measured against Laya 0.3.21 (no /v1/models, no request id, nouls
need instructions, error bodies without a type) and how its answers
compare with Jev's on the demos, including its sensitivity to the
order of the state's fields and its weakness on criteria that need
world knowledge, such as physical plausibility. Update the
TYPESAFE_BASE_URL warnings in the README and the starter page
accordingly.

Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
Label the demo comparison as measured on Laya 0.3.11 and drop its
plausibility row, which disagreed with the controlled 0.3.21 numbers
in the world-knowledge note; the note now carries both phrasings.
List validation status codes among the ITs' expected failures, drop
the CI claim from the intro, and say in the README that the ITs talk
to TYPESAFE_BASE_URL when it is set.

Signed-off-by: Christian Tzolov <christian.tzolov@broadcom.com>
@tzolov
tzolov merged commit bfe85fb into main Sep 29, 2026
1 check passed
@tzolov tzolov added this to the 0.3.0 milestone Sep 29, 2026
@tzolov tzolov added the enhancement New feature or request label Sep 29, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant