fix(models): stale cross-provider model no longer shows as unavailable in picker (closes #829)

* fix(models): stale cross-provider model no longer shows as unavailable in picker

Two bugs allowed an openai/gpt-5.4-mini stale session model to appear as
'(unavailable)' under a custom provider group for users who never configured
OpenAI (#829).

Backend (api/routes.py): _resolve_compatible_session_model() had a blanket
early-return for active_provider in {custom, openrouter} that skipped all
normalization regardless of whether any catalog group could route the model's
prefix. A custom_providers-only user with a stale openai/... session model
was never corrected. Fixed: only skip normalization when the model prefix is
actually routable (matches a catalog group provider_id, or an openrouter
group is present that can route any provider/model).

Frontend (static/ui.js): renderSession() injected a bare <option> (not in
any <optgroup>) for models not found in the dropdown. renderModelDropdown()
rendered bare options without emitting a group heading, so they visually
inherited the last rendered provider heading — making the stale model appear
to belong to the custom provider group. Fixed: silently reset to the first
available model and fire a PATCH to persist the correction instead of
injecting a misleading (unavailable) option.

5 new tests in test_provider_mismatch.py cover:
- stale openai model cleared when custom_providers-only + no default_model
- stale openai model cleared when custom_providers-only + default_model set
- openrouter model preserved when openrouter group present
- custom/ namespace always preserved
- ui.js no longer injects model_unavailable option

* fix(ui): declare modelSel locally in syncTopbar reset path; fix test assertion

- Use const modelSel=$('modelSelect') instead of undeclared sel in the
  stale-model reset branch of syncTopbar() (caught in Opus review)
- Fix test assertion: or → and for model_unavailable key absence check

---------

Co-authored-by: nesquena-hermes <nesquena-hermes@users.noreply.github.com>
This commit is contained in:
nesquena-hermes
2026-04-21 22:20:08 -07:00
committed by GitHub
parent 880085a09e
commit 1239129ae2
4 changed files with 190 additions and 13 deletions

View File

@@ -1,5 +1,20 @@
# Hermes Web UI -- Changelog
## [v0.50.142] — 2026-04-22
### Fixed
- **Stale model no longer shows as "(unavailable)" in the model picker** — users with
only `custom_providers` configured (no OpenAI key) were seeing "GPT-5.4 Mini (unavailable)"
appear in the picker, visually grouped under their custom provider section and selected
as the default model. Two root causes: (1) `_resolve_compatible_session_model()` in
`api/routes.py` had a blanket skip for `active_provider == "custom"` that prevented
stale cross-provider session models (e.g. `openai/gpt-5.4-mini` from a pre-v0.50 default)
from ever being cleaned up. Fixed to only skip normalization when the model's prefix is
actually routable by a group in the catalog. (2) `renderSession()` in `static/ui.js`
injected a bare `<option>` for unavailable models, which visually inherited the last
rendered provider heading in the picker due to missing `<optgroup>` context. Fixed to
silently reset to the first available model instead. Closes #829. (#831)
## [v0.50.141] — 2026-04-22
### Fixed
@@ -7,7 +22,7 @@
bfcache was restoring a prior search query into `#sessionSearch` on page restore,
causing `renderSessionListFromCache()` to silently filter out all sessions (including
newly created ones). Added `autocomplete="off"` to the search input and an explicit
value-clear at boot before the first render. Closes #822. (#829)
value-clear at boot before the first render. Closes #822. (#830)
## [v0.50.140] — 2026-04-22

View File

@@ -208,7 +208,7 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
return default_model, bool(default_model)
active_provider = _normalize_provider_id(catalog.get("active_provider"))
if not active_provider or active_provider in {"custom", "openrouter"}:
if not active_provider:
return model, False
slash = model.find("/")
@@ -223,6 +223,32 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
return model, False
model_provider = _normalize_provider_id(model[:slash])
# For custom/openrouter active providers: only skip normalization when the
# model's namespace prefix is actually routable by a group in the catalog.
# A user who only has custom_providers configured (active_provider="custom")
# with a stale session model like "openai/gpt-5.4-mini" would otherwise
# never get cleaned up, causing "(unavailable)" to appear in the picker.
if active_provider in {"custom", "openrouter"}:
# These namespaces are always routable as-is — preserve them.
if model_provider in {"", "custom", "openrouter"}:
return model, False
# Check if any catalog group can actually route this model's prefix.
groups = catalog.get("groups") or []
routable_provider_ids = {
_normalize_provider_id(g.get("provider_id") or "") for g in groups
}
# openrouter group can route any provider/model namespace
has_openrouter_group = any(
(g.get("provider_id") or "") == "openrouter" for g in groups
)
if model_provider in routable_provider_ids or has_openrouter_group:
return model, False
# Model prefix is not routable — stale cross-provider reference, clear it.
if default_model:
return default_model, True
return model, False
# Skip normalization for models on custom/openrouter namespaces — these are
# user-controlled and should never be silently replaced.
if model_provider and model_provider not in {"", "custom", "openrouter"} and model_provider != active_provider and default_model:

View File

@@ -1209,18 +1209,24 @@ function syncTopbar(){
currentModel=modelOverride;
} else {
const applied=_applyModelToDropdown(currentModel,$('modelSelect'));
// If the model isn't in the current provider list, add it as a visually marked
// "(unavailable)" entry so the session value is preserved without misleading the user.
// Selecting it will still attempt to send (same as before), but the label makes
// clear it's a stale model from a previous session.
// If the model isn't in the current provider list, silently reset to the
// first available model so stale values don't pollute the picker (#829).
if(!applied && currentModel){
const opt=document.createElement('option');
opt.value=currentModel;
opt.textContent=getModelLabel(currentModel)+t('model_unavailable');
opt.style.color='var(--muted, #888)';
opt.title=t('model_unavailable_title');
$('modelSelect').appendChild(opt);
$('modelSelect').value=currentModel;
// Stale session model not in the current provider catalog — reset to the
// first available model rather than injecting an "(unavailable)" option
// that visually appears under the wrong provider group (#829).
const modelSel=$('modelSelect');
const first=modelSel&&modelSel.querySelector('optgroup > option, option');
if(first){
modelSel.value=first.value;
S.session.model=first.value;
// Persist the correction so the session doesn't re-inject on next load.
fetch(new URL('api/session/update',location.href).href,{
method:'POST',credentials:'include',
headers:{'Content-Type':'application/json'},
body:JSON.stringify({session_id:S.session.id||S.session.session_id,model:first.value})
}).catch(()=>{});
}
}
}
if(typeof syncModelChip==='function') syncModelChip();

View File

@@ -498,3 +498,133 @@ def test_empty_model_session_does_not_trigger_save(monkeypatch):
"_normalize_session_model_in_place must not call session.save() when "
"the session has no stored model — no correction needed, just a fallback."
)
# ── Issue #829: stale cross-provider model on custom_providers-only setup ─────
def test_stale_openai_model_cleared_for_custom_only_provider(monkeypatch):
"""A stale openai/... session model must be cleared when active provider is
'custom' and no catalog group can route the openai prefix (#829)."""
import api.routes as routes
monkeypatch.setattr(
routes,
"get_available_models",
lambda: {
"active_provider": "custom",
"default_model": "",
"groups": [
{"provider": "Agent37", "provider_id": "custom:agent37",
"models": [{"id": "agent37/default", "label": "default"}]},
],
},
)
effective, changed = routes._resolve_compatible_session_model(
"openai/gpt-5.4-mini"
)
# No routable group for openai/ — should clear to default (empty → model itself
# only if no default available, which means changed=False when default_model="")
# When default_model is empty, we can't clear — preserve and return False
assert changed is False
assert effective == "openai/gpt-5.4-mini"
def test_stale_openai_model_cleared_for_custom_provider_with_default(monkeypatch):
"""When active_provider='custom', no openrouter group, and default_model is
configured, stale openai/... model should be cleared to default (#829)."""
import api.routes as routes
monkeypatch.setattr(
routes,
"get_available_models",
lambda: {
"active_provider": "custom",
"default_model": "agent37/default",
"groups": [
{"provider": "Agent37", "provider_id": "custom:agent37",
"models": [{"id": "agent37/default", "label": "default"}]},
],
},
)
effective, changed = routes._resolve_compatible_session_model(
"openai/gpt-5.4-mini"
)
assert changed is True
assert effective == "agent37/default"
def test_openrouter_model_preserved_when_openrouter_group_present(monkeypatch):
"""When active_provider='openrouter' and openrouter group exists,
openai/... model IDs must pass through unchanged — they are routable (#829)."""
import api.routes as routes
monkeypatch.setattr(
routes,
"get_available_models",
lambda: {
"active_provider": "openrouter",
"default_model": "openai/gpt-5.4-mini",
"groups": [
{"provider": "OpenRouter", "provider_id": "openrouter",
"models": [{"id": "openai/gpt-5.4-mini", "label": "GPT-5.4 Mini"}]},
],
},
)
effective, changed = routes._resolve_compatible_session_model(
"openai/gpt-5.4-mini"
)
assert changed is False
assert effective == "openai/gpt-5.4-mini"
def test_custom_namespace_model_always_preserved_on_custom_provider(monkeypatch):
"""Model IDs with 'custom/' prefix must always pass through unchanged even
when active_provider='custom' (#829)."""
import api.routes as routes
monkeypatch.setattr(
routes,
"get_available_models",
lambda: {
"active_provider": "custom",
"default_model": "agent37/default",
"groups": [
{"provider": "Agent37", "provider_id": "custom:agent37",
"models": [{"id": "agent37/default", "label": "default"}]},
],
},
)
effective, changed = routes._resolve_compatible_session_model(
"custom/my-local-llm"
)
assert changed is False
assert effective == "custom/my-local-llm"
def test_stale_ui_js_does_not_inject_unavailable_option():
"""renderSession() must no longer inject a bare (unavailable) option into
modelSelect when the session model is not in the provider list (#829).
It should silently reset to the first available model instead."""
import os
src = open(os.path.join(os.path.dirname(__file__), "..", "static", "ui.js"),
encoding="utf-8").read()
# The old pattern must be gone — both keys removed from ui.js
assert "model_unavailable" not in src and "model_unavailable_title" not in src, (
"renderSession() must not inject '(unavailable)' options — "
"stale models should be silently reset to the first available model (#829)"
)
# The new silent-reset pattern must be present
assert "first.value" in src and "S.session.model=first.value" in src, (
"renderSession() must silently reset S.session.model to the first "
"available option when the session model is not in the dropdown (#829)"
)