Skip to content

Agent Zero AI framework

Refactor: Data-driven model search + default presets

keyboardstaff09a74381050b
diff --git a/conf/model_providers.yaml b/conf/model_providers.yaml
index e81091eb2..de037c0f2 100644
--- a/conf/model_providers.yaml
+++ b/conf/model_providers.yaml
@@ -15,11 +15,21 @@
# Optional fields:
# kwargs: A dictionary of extra parameters to pass to LiteLLM.
# This is useful for `api_base`, `extra_headers`, etc.
+#
+# Optional model search fields (used by the Model Configuration plugin):
+# models_endpoint: URL or path for the model listing API.
+# Absolute URL (https://...) is used directly.
+# Relative path (/path) is appended to api_base or default_base.
+# models_format: Response parsing format: "openai" (default), "google", "ollama".
+# models_params: Extra query parameters for the listing request.
+# static_models: Hardcoded model list for providers without a listing API.
+# default_base: Default base URL for local/self-hosted providers.
chat:
a0_venice:
name: Agent Zero API
litellm_provider: openai
+ models_endpoint: "https://api.venice.ai/api/v1/models"
kwargs:
api_base: https://llm.agent-zero.ai/v1
venice_parameters:
@@ -27,15 +37,24 @@ chat:
anthropic:
name: Anthropic
litellm_provider: anthropic
+ models_endpoint: "https://api.anthropic.com/v1/models"
+ models_params:
+ limit: "1000"
cometapi:
name: CometAPI
litellm_provider: cometapi
deepseek:
name: DeepSeek
litellm_provider: deepseek
+ models_endpoint: "https://api.deepseek.com/models"
github_copilot:
name: GitHub Copilot
litellm_provider: github_copilot
+ static_models:
+ - "gpt-4.1"
+ - "gpt-4o"
+ - "gpt-5-mini"
+ - "oswe-vscode-prime"
kwargs:
extra_headers:
"Editor-Version": "vscode/1.85.1"
@@ -44,36 +63,54 @@ chat:
google:
name: Google
litellm_provider: gemini
+ models_endpoint: "/v1beta/models"
+ models_format: "google"
+ models_params:
+ pageSize: "1000"
+ default_base: "https://generativelanguage.googleapis.com"
groq:
name: Groq
litellm_provider: groq
+ models_endpoint: "https://api.groq.com/openai/v1/models"
huggingface:
name: HuggingFace
litellm_provider: huggingface
lm_studio:
name: LM Studio
litellm_provider: lm_studio
+ models_endpoint: "/v1/models"
+ default_base: "http://host.docker.internal:1234"
mistral:
name: Mistral AI
litellm_provider: mistral
+ models_endpoint: "https://api.mistral.ai/v1/models"
moonshot:
name: Moonshot AI
litellm_provider: moonshot
+ models_endpoint: "https://api.moonshot.cn/v1/models"
ollama:
name: Ollama
litellm_provider: ollama
+ models_endpoint: "/api/tags"
+ models_format: "ollama"
+ default_base: "http://host.docker.internal:11434"
openai:
name: OpenAI
litellm_provider: openai
+ models_endpoint: "https://api.openai.com/v1/models"
azure:
name: OpenAI Azure
litellm_provider: azure
+ models_endpoint: "/openai/models"
+ models_params:
+ api-version: "2024-10-21"
bedrock:
name: AWS Bedrock
litellm_provider: bedrock
openrouter:
name: OpenRouter
litellm_provider: openrouter
+ models_endpoint: "https://openrouter.ai/api/v1/models"
kwargs:
extra_headers:
"HTTP-Referer": "https://agent-zero.ai/"
@@ -84,6 +121,7 @@ chat:
sambanova:
name: Sambanova
litellm_provider: sambanova
+ models_endpoint: "https://api.sambanova.ai/v1/models"
venice:
name: Venice.ai
litellm_provider: openai
@@ -94,14 +132,32 @@ chat:
xai:
name: xAI
litellm_provider: xai
+ models_endpoint: "https://api.x.ai/v1/models"
zai:
name: Z.AI
litellm_provider: openai
+ static_models:
+ - "glm-4-plus"
+ - "glm-4-air-250414"
+ - "glm-4-airx"
+ - "glm-4-long"
+ - "glm-4-flashx"
+ - "glm-4-flash-250414"
+ - "glm-4v-plus"
+ - "glm-4v"
+ - "glm-3-turbo"
kwargs:
api_base: https://api.z.ai/api/paas/v4
zai_coding:
name: Z.AI Coding
litellm_provider: openai
+ static_models:
+ - "codegeex-4"
+ - "glm-4-plus"
+ - "glm-4-air-250414"
+ - "glm-4-airx"
+ - "glm-4-flashx"
+ - "glm-4-flash-250414"
kwargs:
api_base: https://api.z.ai/api/coding/paas/v4
other:
diff --git a/plugins/_model_config/api/model_search.py b/plugins/_model_config/api/model_search.py
index 9c9a3d4e0..f2289e15e 100644
--- a/plugins/_model_config/api/model_search.py
+++ b/plugins/_model_config/api/model_search.py
@@ -3,44 +3,6 @@ from helpers.api import ApiHandler, Request, Response
from helpers.providers import get_provider_config
import models
-_CLOUD_ENDPOINTS: dict[str, str] = {
- "openai": "https://api.openai.com/v1/models",
- "anthropic": "https://api.anthropic.com/v1/models",
- "groq": "https://api.groq.com/openai/v1/models",
- "deepseek": "https://api.deepseek.com/models",
- "mistral": "https://api.mistral.ai/v1/models",
- "openrouter": "https://openrouter.ai/api/v1/models",
- "xai": "https://api.x.ai/v1/models",
- "sambanova": "https://api.sambanova.ai/v1/models",
- "moonshot": "https://api.moonshot.cn/v1/models",
- "google": "https://generativelanguage.googleapis.com/v1beta/models",
- "a0_venice": "https://api.venice.ai/api/v1/models",
- "venice": "https://api.venice.ai/api/v1/models",
-}
-
-# Local providers with default base URLs (no auth required).
-_LOCAL_DEFAULTS: dict[str, str] = {
- "ollama": "http://host.docker.internal:11434",
- "lm_studio": "http://host.docker.internal:1234",
-}
-
-# Providers with hardcoded model lists (no listing API available).
-_STATIC_MODELS: dict[str, list[str]] = {
- "github_copilot": [
- "gpt-4.1", "gpt-4o", "gpt-5-mini", "oswe-vscode-prime",
- ],
- "zai": [
- "glm-4-plus", "glm-4-air-250414", "glm-4-airx",
- "glm-4-long", "glm-4-flashx", "glm-4-flash-250414",
- "glm-4v-plus", "glm-4v", "glm-3-turbo",
- ],
- "zai_coding": [
- "codegeex-4",
- "glm-4-plus", "glm-4-air-250414", "glm-4-airx",
- "glm-4-flashx", "glm-4-flash-250414",
- ],
-}
-
# Model name substrings to exclude from litellm fallback results
_LITELLM_EXCLUDE = frozenset({
"dall-e", "gpt-image", "tts", "whisper", "audio",
@@ -58,40 +20,52 @@ class ModelSearch(ApiHandler):
if not provider:
return {"models": []}
- if provider in _STATIC_MODELS:
- all_models = list(_STATIC_MODELS[provider])
+ cfg = self._get_search_config(model_type, provider)
+
+ static = cfg.get("static_models")
+ if static:
+ all_models = list(static)
else:
- provider_cfg = get_provider_config(model_type, provider)
- all_models = await self._fetch_models(provider, provider_cfg, user_api_base) or []
+ all_models = await self._fetch_models(provider, cfg, user_api_base) or []
if not all_models:
- litellm_provider = (provider_cfg or {}).get("litellm_provider", provider)
+ litellm_provider = (cfg or {}).get("litellm_provider", provider)
if litellm_provider == provider:
- all_models = self._litellm_fallback(provider, provider_cfg)
+ all_models = self._litellm_fallback(provider, cfg)
if query:
all_models = [m for m in all_models if query in m.lower()]
return {"models": sorted(all_models)[:50], "provider": provider}
- async def _fetch_models(self, provider: str, cfg: dict | None, user_api_base: str = "") -> list[str] | None:
- api_base = user_api_base or (cfg or {}).get("kwargs", {}).get("api_base", "")
+ @staticmethod
+ def _get_search_config(model_type: str, provider: str) -> dict:
+ """Get provider config with search metadata, falling back to chat config."""
+ cfg = get_provider_config(model_type, provider) or {}
+ if model_type != "chat" and not cfg.get("models_endpoint") and not cfg.get("static_models"):
+ chat_cfg = get_provider_config("chat", provider) or {}
+ merged = dict(cfg)
+ for field in ("models_endpoint", "models_format", "models_params",
+ "static_models", "default_base"):
+ if field in chat_cfg and field not in merged:
+ merged[field] = chat_cfg[field]
+ return merged
+ return cfg
+
+ async def _fetch_models(self, provider: str, cfg: dict, user_api_base: str = "") -> list[str] | None:
api_key = models.get_api_key(provider)
+ api_base = user_api_base or (cfg or {}).get("kwargs", {}).get("api_base", "")
- url, fmt = self._resolve_url(provider, api_base)
+ url, fmt = self._resolve_url(cfg, api_base)
if not url:
return None
headers = self._build_headers(provider, api_key, cfg)
- params: dict[str, str] = {}
- if provider == "google":
- if api_key and api_key != "None":
- params["key"] = api_key
- params["pageSize"] = "1000"
- elif provider == "anthropic":
- params["limit"] = "1000"
- elif provider == "azure":
- params["api-version"] = "2024-10-21"
+ params = dict(cfg.get("models_params", {}) or {})
+
+ # Google uses query-param auth
+ if provider == "google" and api_key and api_key != "None":
+ params.setdefault("key", api_key)
try:
async with httpx.AsyncClient(timeout=10.0) as client:
@@ -105,31 +79,24 @@ class ModelSearch(ApiHandler):
return None
- def _resolve_url(self, provider: str, api_base: str) -> tuple[str | None, str]:
- if provider == "ollama":
- base = api_base or _LOCAL_DEFAULTS.get("ollama", "")
- return (base.rstrip("/") + "/api/tags" if base else None), "ollama"
-
- if provider == "google":
- if api_base:
- return api_base.rstrip("/") + "/models", "google"
- return _CLOUD_ENDPOINTS["google"], "google"
-
- if provider == "azure":
- if not api_base:
- return None, "openai"
- return api_base.rstrip("/") + "/openai/models", "openai"
+ @staticmethod
+ def _resolve_url(cfg: dict, api_base: str) -> tuple[str | None, str]:
+ fmt = cfg.get("models_format", "openai")
+ endpoint = cfg.get("models_endpoint", "")
+ default_base = cfg.get("default_base", "")
- if provider in _CLOUD_ENDPOINTS:
- return _CLOUD_ENDPOINTS[provider], "openai"
+ if endpoint.startswith("http"):
+ return endpoint, fmt
- if api_base:
- return api_base.rstrip("/") + "/models", "openai"
+ base = api_base or default_base
+ if not base:
+ return None, fmt
- if provider in _LOCAL_DEFAULTS:
- return _LOCAL_DEFAULTS[provider] + "/v1/models", "openai"
+ if endpoint:
+ return base.rstrip("/") + endpoint, fmt
- return None, "openai"
+ # Generic fallback: base + /models
+ return base.rstrip("/") + "/models", fmt
def _build_headers(self, provider: str, api_key: str, cfg: dict | None) -> dict[str, str]:
headers: dict[str, str] = {}
diff --git a/plugins/_model_config/default_config.yaml b/plugins/_model_config/default_config.yaml
index 5aae03841..d4a9586d8 100644
--- a/plugins/_model_config/default_config.yaml
+++ b/plugins/_model_config/default_config.yaml
@@ -32,4 +32,26 @@ embedding_model:
browser_http_headers: {}
-model_presets: []
+model_presets:
+ - name: "Efficiency"
+ chat:
+ provider: "openrouter"
+ name: "openai/gpt-5.2-chat"
+ api_key: ""
+ api_base: ""
+ utility:
+ provider: "openrouter"
+ name: "openai/gpt-5-nano"
+ api_key: ""
+ api_base: ""
+ - name: "Intelligence"
+ chat:
+ provider: "openrouter"
+ name: "anthropic/claude-opus-4.6"
+ api_key: ""
+ api_base: ""
+ utility:
+ provider: "openrouter"
+ name: "google/gemini-3-flash-preview"
+ api_key: ""
+ api_base: ""