OpenAI-compatible profile models absent in /model with model_picker_providers set

Summary

When using a custom named provider profile in config.toml (e.g., [providers.ninerouter]) with:

[provider]
default_provider = "ninerouter"
default_model = "gcli/grok-4.5"
model_picker_providers = ["ninerouter"]

The /model endpoint does not display the full static model list from [[providers.ninerouter.models]]. It only shows the currently selected model, expanded into reasoning-effort rows, all labeled:

• PROVIDER: auto
• METHOD: openrouter

Example (only current model visible):

PROVIDER  MODEL                    METHOD
auto      gcli/grok-4.5 (high)     openrouter
auto      gcli/grok-4.5 (low)      openrouter
auto      gcli/grok-4.5 (med)      openrouter
...

Expected:

All configured static models under ninerouter, e.g.:

• cx/gpt-5.6-sol
• cx/gpt-5.6-terra
• cx/gpt-5.6-luna
• gcli/grok-4.5

Environment

• macOS (aarch64)
• jcode remote/shared-server client (remote=true in picker logs)
• Named profile type openai-compatible with model ids that contain /
• Shell also has OPENROUTER_API_KEY set (common; makes the OpenRouter slot look “configured”)

Reproduction Steps

1. Configure a named profile with static models that include / in the id:

[provider]
default_provider = "ninerouter"
default_model = "gcli/grok-4.5"
model_picker_providers = ["ninerouter"]

[providers.ninerouter]
type = "openai-compatible"
base_url = "https://example.gateway/v1"
auth = "bearer"
api_key_env = "NINE_ROUTER_API_KEY"
default_model = "gcli/grok-4.5"
model_catalog = false

[[providers.ninerouter.models]]
id = "cx/gpt-5.6-sol"
[[providers.ninerouter.models]]
id = "cx/gpt-5.6-terra"
[[providers.ninerouter.models]]
id = "cx/gpt-5.6-luna"
[[providers.ninerouter.models]]
id = "gcli/grok-4.5"

2. Ensure credentials for the named profile resolve (env var or env_file under the app config dir).
3. Optionally leave OPENROUTER_API_KEY set in the environment.
4. Start jcode (remote/shared-server path).
5. Open /model.

Actual Behavior

• Picker allowlist keeps only the current model (or only routes that don’t match ninerouter).
• Slash model ids are labeled provider=auto, api_method=openrouter.
• Logs look like:
  EVENT event=model_picker_open by_provider=auto:6 current_model=gcli/grok-4.5 current_provider=ninerouter routes_in=1
• Related earlier failure mode when the profile key was missing:
  Preferred provider ‘ninerouter’ is not configured, using auto-detected default
  Failed to apply default_model ‘…’: NINE_ROUTER_API_KEY not found in environment
  (credential path is separate; the picker labeling bug remains even after the profile works.)

Expected Behavior

• All [[providers.ninerouter.models]] entries appear in /model.
• Routes are labeled with the profile, e.g.:
• provider = "ninerouter"
• api_method = "openai-compatible:ninerouter"
• model_picker_providers = ["ninerouter"] filters to those routes, not to a single current-model OpenRouter auto row.

Context / Root Cause (from investigation)

There appear to be two related gaps:

1. Active named profile static list skipped
   When the named profile is active, catalog building can skip re-emitting [[providers.<name>.models]] under the assumption that the shared OpenRouter-slot runtime already contributes them. With a real OPENROUTER_API_KEY, that slot often contributes a large OpenRouter catalog (auto / openrouter) instead of the profile’s static list.

2. Remote names-only History fallback misclassifies slash model ids
   Initial History often ships model names only (available_model_routes empty for speed). The client rebuilds routes via remote_model_routes_fallback. Any model id containing / is treated as OpenRouter and labeled auto + openrouter.

Start by tracing the model_picker_open path, catalog building for active named profiles, and remote_model_routes_fallback when History contains names only. Reproduce with the ninerouter configuration and inspect how slash-containing model IDs and model_picker_providers are labeled and filtered. Done means all configured static models appear under ninerouter with the expected provider and API method.