refactor: Standardize model provider configuration

conf/model_providers.yaml

@@ -1,55 +1,90 @@
 # Supported model providers for Agent Zero
 # ---------------------------------------
-# Each entry must contain:
-#   id        – identifier used in settings (lower-case, no spaces)
-#   name      – human readable name
-# Optional extra parameters are accepted (api_base, kwargs …)
-# Chat-capable and embedding-capable providers are listed separately as not every
-# provider exposes both kinds of models.
+#
+# Each provider type ("chat", "embedding") contains a mapping of provider IDs
+# to their configurations.
+#
+# The provider ID (e.g., "anthropic") is used:
+#  - in the settings UI dropdowns.
+#  - to construct the environment variable for the API key (e.g., ANTHROPIC_API_KEY).
+#
+# Each provider configuration requires:
+#   name:             Human-readable name for the UI.
+#   litellm_provider: The corresponding provider name in LiteLLM.
+#
+# Optional fields:
+#   kwargs:           A dictionary of extra parameters to pass to LiteLLM.
+#                     This is useful for `api_base`, `extra_headers`, etc.
 
 chat:
-  - id: anthropic
+  anthropic:
     name: Anthropic
-  - id: deepseek
+    litellm_provider: anthropic
+  deepseek:
     name: DeepSeek
-  - id: gemini
+    litellm_provider: deepseek
+  gemini:
     name: Google
-  - id: groq
+    litellm_provider: gemini
+  groq:
     name: Groq
-  - id: huggingface
+    litellm_provider: groq
+  huggingface:
     name: HuggingFace
-  - id: lm_studio
+    litellm_provider: huggingface
+  lm_studio:
     name: LM Studio
-  - id: mistral
+    litellm_provider: lm_studio
+  mistral:
     name: Mistral AI
-  - id: ollama
+    litellm_provider: mistral
+  ollama:
     name: Ollama
-  - id: openai
+    litellm_provider: ollama
+  openai:
     name: OpenAI
-  - id: azure
+    litellm_provider: openai
+  azure:
     name: OpenAI Azure
-  - id: openrouter
+    litellm_provider: azure
+  openrouter:
     name: OpenRouter
-  - id: sambanova
+    litellm_provider: openrouter
+    kwargs:
+      extra_headers:
+        "HTTP-Referer": "https://agent-zero.ai/"
+        "X-Title": "Agent Zero"
+  sambanova:
     name: Sambanova
-  - id: venice
+    litellm_provider: sambanova
+  venice:
     name: Venice
-    api_base: https://api.venice.ai/api/v1
-  - id: other
+    litellm_provider: openai
+    kwargs:
+      api_base: https://api.venice.ai/api/v1
+  other:
     name: Other OpenAI compatible
+    litellm_provider: openai
 
 embedding:
-  - id: huggingface
+  huggingface:
     name: HuggingFace
-  - id: mistral
-    name: Mistral AI
-  - id: lm_studio
+    litellm_provider: huggingface
+  lm_studio:
     name: LM Studio
-  - id: ollama
+    litellm_provider: lm_studio
+  mistral:
+    name: Mistral AI
+    litellm_provider: mistral
+  ollama:
     name: Ollama
-  - id: openai
+    litellm_provider: ollama
+  openai:
     name: OpenAI
-  - id: azure
+    litellm_provider: openai
+  azure:
     name: OpenAI Azure
-  - id: other
-    name: Other OpenAI compatible
+    litellm_provider: azure
+  other:
+    name: Other OpenAI compatible
+    litellm_provider: openai

models.py

@@ -351,6 +351,9 @@ class LocalSentenceTransformerWrapper(Embeddings):
     """Local wrapper for sentence-transformers models to avoid HuggingFace API calls"""
 
     def __init__(self, provider: str, model: str, **kwargs: Any):
+        # Clean common user-input mistakes
+        model = model.strip().strip('"').strip("'")
+
         # Remove the "sentence-transformers/" prefix if present
         if model.startswith("sentence-transformers/"):
             model = model[len("sentence-transformers/") :]
@@ -449,20 +452,37 @@ def _adjust_call_args(provider_name: str, model_name: str, kwargs: dict):
     if provider_name == "other":
         provider_name = "openai"
 
-    # Treat unknown providers that expose a custom OpenAI-compatible endpoint
-    # (i.e. they pass an `api_base` URL) as generic OpenAI providers so that
-    # LiteLLM can route the call correctly. This keeps dedicated providers
-    # such as Azure and OpenRouter unchanged.
-    if kwargs.get("api_base") and provider_name not in (
-        "openai",
-        "azure",
-        "openrouter",
-    ):
-        provider_name = "openai"
-
     return provider_name, model_name, kwargs
 
 
+def _merge_provider_defaults(
+    provider_type: str, original_provider: str, kwargs: dict
+) -> tuple[str, dict]:
+    provider_name = original_provider  # default: unchanged
+    cfg = get_provider_config(provider_type, original_provider)
+    if cfg:
+        provider_name = cfg.get("litellm_provider", original_provider).lower()
+
+        # Extra arguments nested under `kwargs` for readability
+        extra_kwargs = cfg.get("kwargs") if isinstance(cfg, dict) else None  # type: ignore[arg-type]
+        if isinstance(extra_kwargs, dict):
+            for k, v in extra_kwargs.items():
+                kwargs.setdefault(k, v)
+
+        # Copy any additional top-level fields except metadata keys
+        for k, v in cfg.items():
+            if k not in ("id", "name", "value", "litellm_provider", "kwargs"):
+                kwargs.setdefault(k, v)
+
+    # Inject API key based on the *original* provider id if still missing
+    if "api_key" not in kwargs:
+        key = get_api_key(original_provider)
+        if key and key not in ("None", "NA"):
+            kwargs["api_key"] = key
+
+    return provider_name, kwargs
+
+
 def get_model(type: ModelType, provider: str, name: str, **kwargs: Any):
     provider_name = provider.lower()
     if type == ModelType.CHAT:
@@ -476,46 +496,22 @@ def get_model(type: ModelType, provider: str, name: str, **kwargs: Any):
 def get_chat_model(
     provider: str, name: str, **kwargs: Any
 ) -> LiteLLMChatWrapper:
-    provider_name = provider.lower()
-
-    # Merge provider-specific defaults from configuration file
-    cfg = get_provider_config("chat", provider_name)
-    if cfg:
-        extra = {k: v for k, v in cfg.items() if k not in ("id", "name", "value")}
-        for k, v in extra.items():
-            kwargs.setdefault(k, v)
-
-    model = _get_litellm_chat(LiteLLMChatWrapper, name, provider_name, **kwargs)
-    return model
+    orig = provider.lower()
+    provider_name, kwargs = _merge_provider_defaults("chat", orig, kwargs)
+    return _get_litellm_chat(LiteLLMChatWrapper, name, provider_name, **kwargs)
 
 
 def get_browser_model(
     provider: str, name: str, **kwargs: Any
 ) -> BrowserCompatibleChatWrapper:
-    provider_name = provider.lower()
-
-    cfg = get_provider_config("chat", provider_name)
-    if cfg:
-        extra = {k: v for k, v in cfg.items() if k not in ("id", "name", "value")}
-        for k, v in extra.items():
-            kwargs.setdefault(k, v)
-
-    model = _get_litellm_chat(
-        BrowserCompatibleChatWrapper, name, provider_name, **kwargs
-    )
-    return model
+    orig = provider.lower()
+    provider_name, kwargs = _merge_provider_defaults("chat", orig, kwargs)
+    return _get_litellm_chat(BrowserCompatibleChatWrapper, name, provider_name, **kwargs)
 
 
 def get_embedding_model(
     provider: str, name: str, **kwargs: Any
 ) -> LiteLLMEmbeddingWrapper | LocalSentenceTransformerWrapper:
-    provider_name = provider.lower()
-
-    cfg = get_provider_config("embedding", provider_name)
-    if cfg:
-        extra = {k: v for k, v in cfg.items() if k not in ("id", "name", "value")}
-        for k, v in extra.items():
-            kwargs.setdefault(k, v)
-
-    model = _get_litellm_embedding(name, provider_name, **kwargs)
-    return model
+    orig = provider.lower()
+    provider_name, kwargs = _merge_provider_defaults("embedding", orig, kwargs)
+    return _get_litellm_embedding(name, provider_name, **kwargs)

preload.py

@@ -21,11 +21,11 @@ async def preload():
 
         # preload embedding model
         async def preload_embedding():
-            if set["embed_model_provider"] == "HuggingFace":
+            if set["embed_model_provider"].lower() == "huggingface":
                 try:
                     # Use the new LiteLLM-based model system
                     emb_mod = models.get_embedding_model(
-                        "HuggingFace", set["embed_model_name"]
+                        "huggingface", set["embed_model_name"]
                     )
                     emb_txt = await emb_mod.aembed_query("test")
                     return emb_txt

python/helpers/providers.py

@@ -24,20 +24,45 @@ class ProviderManager:
             self._load_providers()
 
     def _load_providers(self):
-        """Loads provider configurations from the YAML file."""
+        """Loads provider configurations from the YAML file and normalises them."""
         try:
             config_path = files.get_abs_path("conf/model_providers.yaml")
             with open(config_path, "r", encoding="utf-8") as f:
-                self._raw = yaml.safe_load(f) or {}
+                raw_yaml = yaml.safe_load(f) or {}
         except (FileNotFoundError, yaml.YAMLError):
-            self._raw = {}
-
-        # Build UI option lists (value / label) from raw data
+            raw_yaml = {}
+
+        # ------------------------------------------------------------
+        # Normalise the YAML so that internally we always work with a
+        # list-of-dicts [{id, name, ...}] for each provider type.  This
+        # keeps existing callers unchanged while allowing the new nested
+        # mapping format in the YAML (id -> { ... }).
+        # ------------------------------------------------------------
+        normalised: Dict[str, List[Dict[str, str]]] = {}
+
+        for p_type, providers in (raw_yaml or {}).items():
+            items: List[Dict[str, str]] = []
+
+            if isinstance(providers, dict):
+                # New format: mapping of id -> config
+                for pid, cfg in providers.items():
+                    entry = {"id": pid, **(cfg or {})}
+                    items.append(entry)
+            elif isinstance(providers, list):
+                # Legacy list format – use as-is
+                items.extend(providers or [])
+
+            normalised[p_type] = items
+
+        # Save raw
+        self._raw = normalised
+
+        # Build UI-friendly option list (value / label)
         self._options = {}
-        for p_type, providers in (self._raw or {}).items():
+        for p_type, providers in normalised.items():
             opts: List[FieldOption] = []
-            for p in providers or []:
-                pid = (p.get("id") or p.get("value") or "").upper()
+            for p in providers:
+                pid = (p.get("id") or p.get("value") or "").lower()
                 name = p.get("name") or p.get("label") or pid
                 if pid:
                     opts.append({"value": pid, "label": name})

python/helpers/settings.py

@@ -500,8 +500,17 @@ def convert_out(settings: Settings) -> SettingsOutput:
     # api keys model section
     api_keys_fields: list[SettingsField] = []
 
-    for provider in get_providers("chat"):
-        api_keys_fields.append(_get_api_key_field(settings, provider["value"].lower(), provider["label"]))
+    # Collect unique providers from both chat and embedding sections
+    providers_seen: set[str] = set()
+    for p_type in ("chat", "embedding"):
+        for provider in get_providers(p_type):
+            pid_lower = provider["value"].lower()
+            if pid_lower in providers_seen:
+                continue
+            providers_seen.add(pid_lower)
+            api_keys_fields.append(
+                _get_api_key_field(settings, pid_lower, provider["label"])
+            )
 
     api_keys_section: SettingsSection = {
         "id": "api_keys",
@@ -993,7 +1002,7 @@ def _write_sensitive_settings(settings: Settings):
 def get_default_settings() -> Settings:
     return Settings(
         version=_get_version(),
-        chat_model_provider="OPENROUTER",
+        chat_model_provider="openrouter",
         chat_model_name="openai/gpt-4.1",
         chat_model_api_base="",
         chat_model_kwargs={"temperature": "0"},
@@ -1003,7 +1012,7 @@ def get_default_settings() -> Settings:
         chat_model_rl_requests=0,
         chat_model_rl_input=0,
         chat_model_rl_output=0,
-        util_model_provider="OPENROUTER",
+        util_model_provider="openrouter",
         util_model_name="openai/gpt-4.1-nano",
         util_model_api_base="",
         util_model_ctx_length=100000,
@@ -1012,13 +1021,13 @@ def get_default_settings() -> Settings:
         util_model_rl_requests=0,
         util_model_rl_input=0,
         util_model_rl_output=0,
-        embed_model_provider="HUGGINGFACE",
+        embed_model_provider="huggingface",
         embed_model_name="sentence-transformers/all-MiniLM-L6-v2",
         embed_model_api_base="",
         embed_model_kwargs={},
         embed_model_rl_requests=0,
         embed_model_rl_input=0,
-        browser_model_provider="OPENROUTER",
+        browser_model_provider="openrouter",
         browser_model_name="openai/gpt-4.1",
         browser_model_api_base="",
         browser_model_vision=True,