fix: route custom provider models dict selections

[Michaelyklam]

Thinking Path

• Issue Clarify and harden Provider/Model source-of-truth across WebUI, CLI, and Gateway #1240 is the broad provider/model source-of-truth umbrella; PR [codex] fix model cache invalidation on config reload #1311 is an older conflicting contributor PR in that area.
• Re-reading [codex] fix model cache invalidation on config reload #1311 against current origin/master showed the cache-invalidation portions have been superseded by the current source-fingerprint cache contract in api/config.py and regression coverage in tests/test_issue1699_model_cache_source_fingerprint.py / tests/test_issue1633_models_cache_version_stamp.py.
• The remaining useful behavior from [codex] fix model cache invalidation on config reload #1311 is narrower: models declared only under custom_providers[].models can appear in the picker, but bare selections of those model IDs were not resolved back to the named custom provider.
• This PR extracts that still-relevant slice into a clean branch from current origin/master instead of pushing to the non-Michael conflicting PR branch.

What Changed

• Updated resolve_model_provider() so named custom_providers match both the singular model field and models dict keys.
• Added a regression test for a custom_providers[].models-only model routing to custom:<name> with the configured base_url.

Why It Matters

• The dropdown path already collects custom_providers[].models dict keys for named custom provider groups.
• Runtime routing now matches that picker behavior, so selecting one of those secondary model IDs does not fall back to the active provider or OpenRouter heuristics.
• This gives maintainers a narrow, mergeable replacement for the non-Michael stale/conflicting [codex] fix model cache invalidation on config reload #1311 slice without trying to close the entire Clarify and harden Provider/Model source-of-truth across WebUI, CLI, and Gateway #1240 umbrella.

Verification

• env -u HERMES_CONFIG_PATH -u HERMES_WEBUI_HOST /home/michael/.hermes/hermes-agent/venv/bin/python -m pytest tests/test_model_resolver.py::test_custom_provider_models_dict_routes_to_named_custom_provider -q
  • First run before implementation failed as expected: provider resolved to xiaomi instead of custom:litellm-proxy.
  • Final run passed.
• env -u HERMES_CONFIG_PATH -u HERMES_WEBUI_HOST /home/michael/.hermes/hermes-agent/venv/bin/python -m pytest tests/test_model_resolver.py tests/test_issue1699_model_cache_source_fingerprint.py tests/test_issue1633_models_cache_version_stamp.py tests/test_ttl_cache.py tests/test_pr1339_fallback_providers_list.py -q
  • 51 passed in 48.98s
• env -u HERMES_CONFIG_PATH -u HERMES_WEBUI_HOST /home/michael/.hermes/hermes-agent/venv/bin/python -m pytest tests/ -q
  • 4597 passed, 2 skipped, 3 xpassed, 1 warning, 8 subtests passed in 378.87s
• git diff --check

Risks / Follow-ups

• This intentionally does not attempt to close Clarify and harden Provider/Model source-of-truth across WebUI, CLI, and Gateway #1240 wholesale; it only reconciles one concrete stale PR slice.
• PR [codex] fix model cache invalidation on config reload #1311 still needs maintainer cleanup/closure after this lands because its cache-invalidation changes are superseded and its branch is conflicting.
• No UI screenshots: server-side routing only; no interface or interaction flow changed.

Model Used

• OpenAI Codex gpt-5.5 via Hermes Agent CLI.
• Tool use: GitHub CLI, git worktree, pytest, and file-edit tools.

[nesquena-hermes]

Thanks @Michaelyklam — this shipped in v0.51.12 (commit 34f2243) as part of a 3-PR full-sweep batch release. Stage rebased your branch onto current master, ran the full pre-release gate (4632 pytest, browser tests, Opus advisor verdict SHIP), and merged via release PR #1755.

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 https://get-hermes.ai/ and on existing installs after git pull + restart.

Release notes: https://github.com/nesquena/hermes-webui/releases/tag/v0.51.12