Refactor config file and Implement OpenRouter embedding function

src/bandai/config/__init__.py

@@ -1,18 +1,22 @@
-from .config import (
+from ._constants import (  # noqa: F401
     _MAX_REVIEW_ITERATIONS,
-    EnvOverrides,
+    PROVIDERS,
+    LLMProfile,
+    ProviderProfile,
     clear_caches,
+)
+from ._env import (  # noqa: F401
+    EnvOverrides,
     get_active_provider,
     get_api_key,
+)
+from .llm import get_llm  # noqa: F401
+from .embedder import (  # noqa: F401
     get_embedder,
-    get_llm,
-    get_memory,
-    validate_config,
-    PROVIDERS,
-    ProviderProfile,
-    LLMProfile,
 )
-from .portals import (
+from .memory import get_memory  # noqa: F401
+from .validation import validate_config  # noqa: F401
+from .portals import (  # noqa: F401
     BANDI_PORTALS,
     PORTAL_WEIGHTS,
     PortalConfig,

src/bandai/config/_constants.py

@@ -0,0 +1,104 @@
+from __future__ import annotations
+
+import logging
+import os
+from typing import TYPE_CHECKING
+
+from pydantic import BaseModel, Field
+
+if TYPE_CHECKING:
+    pass
+
+log = logging.getLogger(__name__)
+logging.getLogger("opentelemetry.attributes").setLevel(logging.ERROR)
+
+
+# Provider Profile Models
+
+
+class LLMProfile(BaseModel):
+    """Configuration for a single LLM model slot (main or fast)."""
+
+    model: str
+    temperature: float = Field(default=0.3, ge=0.0, le=2.0)
+    max_tokens: int = Field(default=4096, ge=1, le=131072)
+
+
+class ProviderProfile(BaseModel):
+    """Configuration for a complete LLM provider."""
+
+    name: str
+    description: str = ""
+    base_url: str
+    main: LLMProfile
+    fast: LLMProfile
+    env_key: str = "API_KEY"
+
+    @property
+    def main_model(self) -> str:
+        """Full model identifier in provider/model format."""
+        if "/" in self.main.model:
+            return self.main.model
+        return f"{self.name}/{self.main.model}"
+
+    @property
+    def fast_model(self) -> str:
+        """Full model identifier in provider/model format."""
+        if "/" in self.fast.model:
+            return self.fast.model
+        return f"{self.name}/{self.fast.model}"
+
+
+# Built-in Provider Profiles
+
+PROVIDERS: dict[str, ProviderProfile] = {
+    "openrouter": ProviderProfile(
+        name="openrouter",
+        description="OpenRouter - multi-provider routing (supports OpenAI, Anthropic, Google, etc.)",
+        base_url="https://openrouter.ai/api/v1",
+        main=LLMProfile(model="anthropic/claude-sonnet-4-20250514", temperature=0.3, max_tokens=4096),
+        fast=LLMProfile(model="openai/gpt-4o-mini", temperature=0.3, max_tokens=4096),
+        env_key="OPENROUTER_API_KEY",
+    ),
+    "anthropic": ProviderProfile(
+        name="anthropic",
+        description="Anthropic - direct API access",
+        base_url="https://api.anthropic.com/v1",
+        main=LLMProfile(model="claude-sonnet-4-20250514", temperature=0.3, max_tokens=4096),
+        fast=LLMProfile(model="claude-haiku-3-5-20241022", temperature=0.3, max_tokens=4096),
+        env_key="ANTHROPIC_API_KEY",
+    ),
+    "openai": ProviderProfile(
+        name="openai",
+        description="OpenAI - direct API access",
+        base_url="https://api.openai.com/v1",
+        main=LLMProfile(model="gpt-4o", temperature=0.3, max_tokens=4096),
+        fast=LLMProfile(model="gpt-4o-mini", temperature=0.3, max_tokens=4096),
+        env_key="OPENAI_API_KEY",
+    ),
+    "ollama": ProviderProfile(
+        name="ollama",
+        description="Ollama - local LLM server",
+        base_url="http://localhost:11434/v1",
+        main=LLMProfile(model="llama3", temperature=0.3, max_tokens=4096),
+        fast=LLMProfile(model="llama3", temperature=0.3, max_tokens=4096),
+        env_key="",
+    ),
+}
+
+
+# Review Loop Limits
+
+_MAX_REVIEW_ITERATIONS = int(os.getenv("MAX_REVIEW_ITERATIONS", "5"))
+
+
+# Cache Management (testing)
+
+
+def clear_caches() -> None:
+    """Clear all LLM/embedder caches.  Useful in tests."""
+    from bandai.config.llm import get_llm  # noqa: E402
+    from bandai.config.embedder import get_embedder  # noqa: E402
+
+    get_llm.cache_clear()
+    get_embedder.cache_clear()

src/bandai/config/_env.py

@@ -0,0 +1,81 @@
+from __future__ import annotations
+
+import os
+
+from pydantic import BaseModel, Field
+
+from bandai.config._constants import PROVIDERS, ProviderProfile  # noqa: E402
+
+# Environment Override Model
+
+
+class EnvOverrides(BaseModel):
+    """Parsed LLM environment variable overrides."""
+
+    model: str | None = None
+    temperature: float | None = Field(default=None, ge=0.0, le=2.0)
+    max_tokens: int | None = Field(default=None, ge=1, le=131072)
+
+    @classmethod
+    def from_env(cls, prefix: str) -> EnvOverrides:
+        """Read overrides for *prefix* (MAIN_ or FAST_)."""
+
+        def _float(key: str) -> float | None:
+            raw = os.getenv(key)
+            if raw is None:
+                return None
+            try:
+                return float(raw)
+            except ValueError:
+                return None
+
+        def _int(key: str) -> int | None:
+            raw = os.getenv(key)
+            if raw is None:
+                return None
+            try:
+                return int(raw)
+            except ValueError:
+                return None
+
+        return cls(
+            model=os.getenv(f"{prefix}MODEL") or None,
+            temperature=_float(f"{prefix}TEMPERATURE"),
+            max_tokens=_int(f"{prefix}MAX_TOKENS"),
+        )
+
+
+# Active Configuration
+
+
+def get_active_provider() -> ProviderProfile:
+    """Return the active provider profile based on LLM_PROVIDER env var."""
+    provider_name = os.getenv("LLM_PROVIDER", "openrouter")
+    if provider_name not in PROVIDERS:
+        available = ", ".join(sorted(PROVIDERS.keys()))
+        raise ValueError(f"Unknown LLM_PROVIDER '{provider_name}'. " f"Available providers: {available}")
+    return PROVIDERS[provider_name]
+
+
+def get_api_key() -> str:
+    """Return the API key for the active provider."""
+    provider = get_active_provider()
+
+    # Provider-specific key takes priority
+    if provider.env_key:
+        key = os.getenv(provider.env_key, "")
+        if key and key != f"your_{provider.env_key.lower()}_here":
+            return key
+
+    # Generic fallback
+    key = os.getenv("API_KEY", "")
+    if key and key != "your_api_key_here":
+        return key
+
+    available_msg = ""
+    if provider.env_key:
+        available_msg = f" Set {provider.env_key} or API_KEY in your .env file."
+    else:
+        available_msg = " No API key required for this provider."
+
+    raise ValueError(f"No API key configured for provider '{provider.name}'.{available_msg}")