feat(agent): add Upstage Solar as a model provider

Adds Upstage Solar as a bundled model-provider plugin. Solar exposes an
OpenAI-compatible chat-completions endpoint at https://api.upstage.ai/v1, so
the generic chat_completions transport handles request/response/streaming/tool
calls — the profile is the core integration.

Provider registration (Upstage isn't in models.dev, so each registry that does
not auto-wire from the plugin layer needs an explicit entry — same pattern as
nvidia/gmi):
- plugins/model-providers/upstage/: UpstageProfile + plugin.yaml. Picker default
  and offline catalog list only the agentic Solar Pro models, led by `solar-pro`
  (rolling alias for the latest Pro). default_aux_model empty so aux tasks use
  the main model. `solar` alias. UPSTAGE_BASE_URL overrides the host.
- hermes_cli/providers.py: HERMES_OVERLAYS + label + `solar` alias, so
  resolve_provider_full('upstage') resolves (without this, an explicit
  `provider: upstage` in config was dropped and fell through to auto-detect).
- hermes_cli/auth.py: PROVIDER_REGISTRY entry + `solar` alias, so `hermes
  doctor` / resolve_provider recognise upstage (the static-registry path the
  lazy profile-extension doesn't reliably cover at validation time).
- hermes_cli/models.py: CANONICAL_PROVIDERS entry places Upstage Solar in the
  curated picker order (above the auto-appended `custom`).
- agent/model_metadata.py: context-window fallbacks (/v1/models omits
  context_length); `solar-pro` carries the 128K Pro context as the catch-all.

Reasoning: UpstageProfile.build_api_kwargs_extras wires Solar's top-level
`reasoning_effort` (low|medium|high; xhigh/max→high). Reasoning-capable families
are solar-pro* and solar-open*; solar-mini/syn-pro never receive it. Defaults ON
at medium when unset (matches the /reasoning "medium (default)" label);
`/reasoning none` disables; explicit/saved settings are honored. No
reasoning_content echo handling needed (unlike DeepSeek/Kimi).

Web dashboard:
- web/src/pages/EnvPage.tsx: add an "Upstage Solar" provider group so
  UPSTAGE_API_KEY / UPSTAGE_BASE_URL appear under LLM Providers (not "Other").

Docs/tests:
- .env.example: documents UPSTAGE_API_KEY / UPSTAGE_BASE_URL.
- tests: profile wiring, reasoning_effort mapping (pro/open/mini, efforts,
  disabled, default-on), provider-resolver regression (resolve_provider_full /
  get_provider / solar alias / overlay), `solar-pro` default.

Testing: pytest tests/providers tests/plugins/model_providers
tests/hermes_cli/test_upstage_provider.py tests/run_agent/test_provider_parity.py
tests/hermes_cli/test_api_key_providers.py; ruff clean. Verified end-to-end:
`hermes doctor` shows "Upstage Solar", and live chat works via both
`--provider upstage` and `--provider solar`. Reasoning wire format per
https://console.upstage.ai/api/docs/for-agents/raw. Platforms tested: macOS.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

plugins/model-providers/upstage/__init__.py

@@ -0,0 +1,106 @@
+"""Upstage Solar provider profile."""
+
+from typing import Any
+
+from providers import register_provider
+from providers.base import ProviderProfile
+
+
+# Model-name markers for Solar families that accept ``reasoning_effort``.
+# Substring match (not startswith) so dated variants like
+# ``solar-pro3-250127`` and ``solar-open-...`` are covered too.
+_REASONING_MODEL_MARKERS = ("solar-pro", "solar-open")
+
+# When the user hasn't picked a reasoning effort, Hermes passes
+# reasoning_config=None. Solar's own server default is "minimal" (reasoning
+# off), which is the wrong default for an agentic workload. We default reasoning
+# ON at this effort — matching the "medium (default)" that Hermes' /reasoning
+# panel shows for an unset config, so the displayed default and the real wire
+# value agree. An explicit saved setting or a `/reasoning <level>` change is
+# always honored over this default; `/reasoning none` disables it.
+_DEFAULT_REASONING_EFFORT = "medium"
+
+
+def _model_supports_reasoning(model: str | None) -> bool:
+    """Solar reasoning-capable models.
+
+    The Solar Pro family (``solar-pro``, ``solar-pro2``, ``solar-pro3`` and
+    dated variants like ``solar-pro3-250127``) and the Solar Open family
+    (``solar-open*``) accept ``reasoning_effort``. ``solar-mini`` / ``syn-pro``
+    ignore the parameter, so we don't send it.
+    """
+    m = (model or "").strip().lower()
+    return any(marker in m for marker in _REASONING_MODEL_MARKERS)
+
+
+class UpstageProfile(ProviderProfile):
+    """Upstage Solar — top-level ``reasoning_effort`` control.
+
+    Solar Pro/Open expose reasoning through a top-level ``reasoning_effort``
+    field (``minimal`` | ``low`` | ``medium`` | ``high``), mirroring OpenAI's
+    shape. Unlike DeepSeek/Kimi it does NOT require echoing ``reasoning_content``
+    back on later turns, so only the request field needs wiring. We emit at most
+    ``low`` | ``medium`` | ``high`` — the explicit values both Solar Pro 2 and
+    Pro 3 accept.
+
+    Default-on: Solar's own server default is ``minimal`` (off), but for an
+    agentic workload we default reasoning ON (``_DEFAULT_REASONING_EFFORT``)
+    when the user hasn't picked an effort. The user can still set any level or
+    turn it off with ``/reasoning none``.
+    """
+
+    def build_api_kwargs_extras(
+        self, *, reasoning_config: dict | None = None, model: str | None = None, **context
+    ) -> tuple[dict[str, Any], dict[str, Any]]:
+        top_level: dict[str, Any] = {}
+
+        # solar-mini / syn-pro ignore reasoning_effort — send nothing.
+        if not _model_supports_reasoning(model):
+            return {}, top_level
+
+        # Unset (reasoning_config is None) → default reasoning ON for agents.
+        if not reasoning_config or not isinstance(reasoning_config, dict):
+            return {}, {"reasoning_effort": _DEFAULT_REASONING_EFFORT}
+
+        # Explicitly disabled (`/reasoning none`) → omit the field so Solar
+        # applies its own default (minimal = off).
+        if reasoning_config.get("enabled") is False:
+            return {}, top_level
+
+        # Map Hermes' effort vocabulary onto Solar's accepted set. xhigh/max
+        # collapse to high (Solar's strongest). minimal → off (omit). An
+        # enabled request with no recognised effort uses the default effort.
+        effort = (reasoning_config.get("effort") or "").strip().lower()
+        mapped = {
+            "minimal": None,
+            "low": "low",
+            "medium": "medium",
+            "high": "high",
+            "xhigh": "high",
+            "max": "high",
+        }.get(effort, _DEFAULT_REASONING_EFFORT)
+
+        if mapped:
+            top_level["reasoning_effort"] = mapped
+        return {}, top_level
+
+
+upstage = UpstageProfile(
+    name="upstage",
+    aliases=("solar",),
+    display_name="Upstage Solar",
+    description="Upstage (Solar API)",
+    signup_url="https://console.upstage.ai/api-keys",
+    env_vars=("UPSTAGE_API_KEY", "UPSTAGE_BASE_URL"),
+    base_url="https://api.upstage.ai/v1",
+    auth_type="api_key",
+    # default_aux_model left empty → auxiliary side tasks use the main model.
+    # entry [0] is the setup default. solar-pro is a rolling alias for the
+    # latest Solar Pro, so the default tracks the current flagship.
+    fallback_models=(
+        "solar-pro",
+        "solar-pro3",
+    ),
+)
+
+register_provider(upstage)

plugins/model-providers/upstage/plugin.yaml

@@ -0,0 +1,5 @@
+name: upstage-provider
+kind: model-provider
+version: 1.0.0
+description: Upstage (Solar API)
+author: Upstage AI