Say what the provider said, bound silent streams, ai-chat 0.5 - #344
Merged
Merged
Conversation
…t 0.5
The availability banner explained every [unavailable] mark as a key on
an 8-hour lock schedule. Measured against a provider whose models were
marked, the cause was the provider refusing the model names outright,
streamed and not ("Requested model is not available and no compliant
same-model variant was found"), which no unlock would fix. The mark was
right and the explanation wrong. Each marked model now carries the
provider's own sentence, unwrapped from the gateway's nested error
bodies, and the banner quotes it; a lock still gets the unlock step.
A stream that opens and then sends nothing had no limit other than the
client's own abort. Each read now has one, reset by every chunk and set
at five minutes, since some providers buffer a whole answer and take
about a minute per call.
ai-chat moves to 0.5.0 and @parallelworks/ui to 0.16, its new floor.
The attachments list now pages by cursor and reports hasMore; the
adapter maps the server's offset to that contract. The chat renders and
lists models with no page errors in a headless check. The change log is
regenerated, since the last two entries never reached the default
branch.
The earlier probe change (v1.60) removed the token cap on the claim that
a cap-rejecting family was being marked; its live verification had hit
the wrong gateway, and the marks it was meant to clear had this other
cause. The cap removal stays, since a cap was never needed for a probe.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Banner. Every
[unavailable]mark was explained as a key on an 8-hour lock schedule. Measured against a provider whose models were marked: the provider refused the model names outright, streamed (502) and non-streamed (masked 400), with Requested model is not available and no compliant same-model variant was found. The mark was correct; the explanation was wrong. Marked models now carry the provider's own sentence (providerReason, which unwraps the gateway's nested error bodies) and the banner quotes it. A lock still gets the unlock step.Correction. #340 claimed live verification that removing the probe's token cap cleared the marks. That check hit the wrong gateway (
Provider owner not found), and the marks had this other cause. Removing the cap stays, since a probe never needed one.Silent streams. A stream that opened and sent nothing had no limit but the client's abort. Each read now has one (
STUDIO_STREAM_IDLE_MS, default 300 s), reset by every chunk; some providers buffer a whole answer and take about a minute per call.ai-chat 0.5.0 with
@parallelworks/ui^0.16 (its peer floor). Attachments paging moved to a cursor contract withhasMore; the adapter maps the server offset to it (tested). Headless smoke: chat canvas, composer, and picker render, 19 models listed, no page errors.Also regenerates CHANGELOG.md; the v1.62 and v1.63 entries never reached main. 189 server and 15 web tests pass.