fix(config): custom provider + :free/:beta/:thinking suffix mis-resolution (#1776)

[Sanjays2402]

Summary

Closes #1776.

PR #1762 fixed the rsplit grammar collision for plain @openrouter:model:free qualifiers, but skipped the fallback whenever the provider hint started with custom: on the assumption that custom providers route directly. That left @custom:my-key:some-model:free broken — issue body and the maintainer's confirmation comment trace the exact failure mode.

Approach

Adopted the maintainer's recommended approach (no allowlist needed). Custom hints carry exactly one segment after custom: (constructed at api/config.py:1363 as \"custom:\" + entry_name.lower().replace(\" \", \"-\")), so any rsplit result of custom:<a>:<b> with bare model <c> has eaten a model segment. Peel it back with a second rsplit and prepend to the bare model.

provider_hint, bare_model = inner.rsplit(":", 1)
if provider_hint.startswith("custom:") and provider_hint.count(":") >= 2:
    provider_hint, extra = provider_hint.rsplit(":", 1)
    bare_model = f"{extra}:{bare_model}"
elif (provider_hint not in _PROVIDER_MODELS
        and provider_hint not in _PROVIDER_DISPLAY
        and not provider_hint.startswith("custom:")):
    provider_hint, bare_model = inner.split(":", 1)

Robust for :free / :beta / :thinking / :preview / any future OpenRouter suffix — no allowlist to keep in sync.

Resolution matrix (verified by tests)

Input	provider	model
@custom:my-key:some-model:free	custom:my-key	some-model:free ✅
@custom:my-key:some-model:beta	custom:my-key	some-model:beta ✅
@custom:my-key:some-model:thinking	custom:my-key	some-model:thinking ✅
@custom:my-key:some-model:preview	custom:my-key	some-model:preview ✅
@custom:my-key:org/model:free	custom:my-key	org/model:free ✅
@custom:my-key:some-model (no suffix)	custom:my-key	some-model ✅ (existing test)
@openrouter:tencent/hy3-preview:free	openrouter	tencent/hy3-preview:free ✅ (existing test)
@anthropic:claude-sonnet-4.6	anthropic	claude-sonnet-4.6 ✅ (existing test)

Tests

$ python -m pytest tests/test_resolve_model_provider_free_suffix.py tests/test_issue1228_model_picker_duplicate_ids.py -v
============================== 29 passed in 2.77s ==============================

• 5 new regression tests in tests/test_resolve_model_provider_free_suffix.py covering the issue matrix
• All 7 existing OpenRouter model tencent/hy3-preview:free fails with API key not found error in WebUI #1744 tests still pass
• All 17 bug(models): session model picker loses provider identity when multiple providers expose the same model ID #1228 tests (related custom-provider routing) still pass

Open question (left for separate issue)

Per the maintainer's comment: entry_name.lower().replace(\" \", \"-\") doesn't strip colons, so a user who literally names a custom provider \"foo:bar\" would still produce a custom:foo:bar slug that this fix would mis-handle. Probably worth a separate input-validation issue to forbid : in custom_provider names rather than complicating this resolver. Happy to file a follow-up if useful.

[nesquena-hermes]

Thanks @Sanjays2402 — this shipped in v0.51.18 (commit a8de4e7) as part of a 5-PR full-sweep batch. Stage rebased your branch onto current master, ran the full pre-release gate (4733 pytest, browser tests, Opus advisor verdict SHIP all 5), and merged via release PR #1799.

GitHub didn't auto-close because the merge commit only references the squash-merged stage branch, not your fork's commit directly — closing manually for hygiene.

Live now on existing installs after git pull + restart.

Release notes: https://github.com/nesquena/hermes-webui/releases/tag/v0.51.18