ai_lib/provider/
configs.rs

1use super::config::ProviderConfig;
2
3/// Predefined provider configurations for multiple AI services
4///
5/// Each provider configuration includes default models for chat and optional multimodal support.
6/// Developers can override these defaults or build custom model managers for advanced use cases.
7pub struct ProviderConfigs;
8
9impl ProviderConfigs {
10    pub fn groq() -> ProviderConfig {
11        ProviderConfig::openai_compatible(
12            "https://api.groq.com/openai/v1",
13            "GROQ_API_KEY",
14            "llama-3.1-8b-instant", // Default chat model - fast and cost-effective
15            Some("llama-3.2-11b-vision"), // Multimodal / vision-capable model
16        )
17    }
18
19    pub fn openai() -> ProviderConfig {
20        ProviderConfig::openai_compatible(
21            "https://api.openai.com/v1",
22            "OPENAI_API_KEY",
23            "gpt-3.5-turbo", // Default chat model
24            Some("gpt-4o"),  // Multimodal model
25        )
26    }
27
28    pub fn deepseek() -> ProviderConfig {
29        ProviderConfig::openai_compatible(
30            "https://api.deepseek.com/v1",
31            "DEEPSEEK_API_KEY",
32            "deepseek-chat", // Default chat model
33            None,            // No multimodal support yet
34        )
35    }
36
37    pub fn ollama() -> ProviderConfig {
38        // Ollama is commonly run locally and is OpenAI-compatible in many setups.
39        // Allow developers to override the local base URL via OLLAMA_BASE_URL.
40        // Default remains the common local address used by Ollama.
41        let base_url = std::env::var("OLLAMA_BASE_URL")
42            .unwrap_or_else(|_| "http://localhost:11434/api".to_string());
43
44        // Ollama typically doesn't require an API key for local installs. We keep
45        // the env variable name for parity, but it's optional for users.
46        ProviderConfig::openai_compatible(
47            &base_url,
48            "OLLAMA_API_KEY",
49            "llama3-8b", // Default chat model - common local model
50            None,        // No multimodal support yet
51        )
52    }
53
54    /// xAI / Grok configuration - OpenAI-compatible hosted offering
55    pub fn xai_grok() -> ProviderConfig {
56        let base_url = std::env::var("GROK_BASE_URL")
57            .unwrap_or_else(|_| "https://api.grok.com/openai/v1".to_string());
58        ProviderConfig::openai_compatible(
59            &base_url,
60            "GROK_API_KEY",
61            "grok-beta", // Default chat model
62            None,        // No multimodal support yet
63        )
64    }
65
66    /// Azure OpenAI configuration - highly compatible but often uses resource-specific base URL
67    pub fn azure_openai() -> ProviderConfig {
68        // Expect developers to set AZURE_OPENAI_BASE_URL to their resource endpoint
69        let base_url = std::env::var("AZURE_OPENAI_BASE_URL")
70            .unwrap_or_else(|_| "https://api.openai.azure.com/v1".to_string());
71        ProviderConfig::openai_compatible(
72            &base_url,
73            "AZURE_OPENAI_API_KEY",
74            "gpt-35-turbo", // Default Azure OpenAI model
75            Some("gpt-4o"), // Multimodal model
76        )
77    }
78
79    /// Hugging Face Inference API - configured to reuse generic adapter; may need adjustments per model
80    pub fn huggingface() -> ProviderConfig {
81        // Hugging Face inference API default (used for inference calls)
82        let base_url = std::env::var("HUGGINGFACE_BASE_URL")
83            .unwrap_or_else(|_| "https://api-inference.huggingface.co".to_string());
84
85        use std::collections::HashMap;
86        let mut headers = HashMap::new();
87        headers.insert("Content-Type".to_string(), "application/json".to_string());
88
89        // Hugging Face model listing is served from the Hub API under huggingface.co.
90        // Use an absolute models endpoint so we can query the Hub independently of the
91        // inference base URL (inference and hub are different services).
92        ProviderConfig {
93            base_url: base_url.clone(),
94            api_key_env: "HUGGINGFACE_API_KEY".to_string(),
95            chat_endpoint: "/models/{model}:predict".to_string(), // placeholder; per-model inference often requires model in path
96            chat_model: "microsoft/DialoGPT-medium".to_string(),  // Default chat model
97            multimodal_model: None,                               // No multimodal support yet
98            upload_endpoint: None,
99            upload_size_limit: None,
100            models_endpoint: Some("https://huggingface.co/api/models".to_string()),
101            headers,
102            field_mapping: crate::provider::config::FieldMapping {
103                messages_field: "messages".to_string(),
104                model_field: "model".to_string(),
105                role_mapping: {
106                    let mut role_mapping = std::collections::HashMap::new();
107                    role_mapping.insert("System".to_string(), "system".to_string());
108                    role_mapping.insert("User".to_string(), "user".to_string());
109                    role_mapping.insert("Assistant".to_string(), "assistant".to_string());
110                    role_mapping
111                },
112                response_content_path: "choices.0.message.content".to_string(),
113            },
114        }
115    }
116
117    /// Together AI - OpenAI-compatible chat API
118    pub fn together_ai() -> ProviderConfig {
119        let base_url = std::env::var("TOGETHER_BASE_URL")
120            .unwrap_or_else(|_| "https://api.together.ai".to_string());
121        ProviderConfig::openai_compatible(
122            &base_url,
123            "TOGETHER_API_KEY",
124            "meta-llama/Llama-3-8b-chat-hf", // Default chat model
125            None,                            // No multimodal support yet
126        )
127    }
128
129    /// Groq configuration - proving OpenAI compatibility
130    pub fn groq_as_generic() -> ProviderConfig {
131        ProviderConfig::openai_compatible(
132            "https://api.groq.com/openai/v1",
133            "GROQ_API_KEY",
134            "llama-3.1-8b-instant", // Default chat model
135            None,                   // No multimodal support yet
136        )
137    }
138
139    /// Qwen / Tongyi Qianwen (Alibaba Cloud) - OpenAI-compatible mode
140    /// Uses DASHSCOPE_API_KEY and optional DASHSCOPE_BASE_URL to override the base URL.
141    pub fn qwen() -> ProviderConfig {
142        // Default to Alibaba DashScope public base URL and use the generation endpoint.
143        let base_url = std::env::var("DASHSCOPE_BASE_URL")
144            .unwrap_or_else(|_| "https://dashscope.aliyuncs.com".to_string());
145
146        // DashScope generation endpoint example: /api/v1/services/aigc/text-generation/generation
147        let mut cfg = ProviderConfig::openai_compatible(
148            &base_url,
149            "DASHSCOPE_API_KEY",
150            "qwen-turbo", // Default chat model
151            None,         // No multimodal support yet
152        );
153        cfg.chat_endpoint = "/api/v1/services/aigc/text-generation/generation".to_string();
154        cfg
155    }
156
157    /// Baidu Wenxin (ERNIE) - OpenAI compatible mode via Qianfan/Console configuration
158    /// Environment variables: BAIDU_WENXIN_BASE_URL (optional), BAIDU_WENXIN_API_KEY, BAIDU_WENXIN_SECRET
159    pub fn baidu_wenxin() -> ProviderConfig {
160        let base_url = std::env::var("BAIDU_WENXIN_BASE_URL")
161            .unwrap_or_else(|_| "https://aip.baidubce.com".to_string());
162
163        // Baidu Wenxin example path (model-specific):
164        // /rpc/2.0/ai_custom/v1/wenxinworkshop/chat/{model_name}
165        let mut cfg = ProviderConfig::openai_compatible(
166            &base_url,
167            "BAIDU_WENXIN_API_KEY",
168            "ernie-3.5", // Default chat model
169            None,        // No multimodal support yet
170        );
171        cfg.chat_endpoint = "/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/{model}".to_string();
172        // Baidu may use different upload endpoints; keep defaults but don't assume model listing.
173        cfg.models_endpoint = None;
174        cfg
175    }
176
177    /// Tencent Hunyuan - Tencent Cloud OpenAI compatible endpoint
178    /// Environment variables: TENCENT_HUNYUAN_BASE_URL (optional), TENCENT_HUNYUAN_API_KEY
179    pub fn tencent_hunyuan() -> ProviderConfig {
180        let base_url = std::env::var("TENCENT_HUNYUAN_BASE_URL")
181            .unwrap_or_else(|_| "https://hunyuan.tencentcloudapi.com".to_string());
182
183        // Tencent Hunyuan example uses an Action query parameter: ?Action=ChatCompletions
184        let mut cfg = ProviderConfig::openai_compatible(
185            &base_url,
186            "TENCENT_HUNYUAN_API_KEY",
187            "hunyuan-standard", // Default chat model
188            None,               // No multimodal support yet
189        );
190        cfg.chat_endpoint = "/?Action=ChatCompletions".to_string();
191        cfg.models_endpoint = None;
192        cfg
193    }
194
195    /// iFlytek Spark - OpenAI compatible endpoint example
196    /// Environment variables: IFLYTEK_BASE_URL (optional), IFLYTEK_API_KEY
197    pub fn iflytek_spark() -> ProviderConfig {
198        let base_url = std::env::var("IFLYTEK_BASE_URL")
199            .unwrap_or_else(|_| "https://api.xf-yun.com".to_string());
200        let mut cfg = ProviderConfig::openai_compatible(
201            &base_url,
202            "IFLYTEK_API_KEY",
203            "spark-v3.0", // Default chat model
204            None,         // No multimodal support yet
205        );
206        cfg.chat_endpoint = "/v1/chat".to_string();
207        cfg
208    }
209
210    /// Moonshot (Kimi) - OpenAI compatible endpoint
211    /// Environment variables: MOONSHOT_BASE_URL (optional), MOONSHOT_API_KEY
212    pub fn moonshot() -> ProviderConfig {
213        let base_url = std::env::var("MOONSHOT_BASE_URL")
214            .unwrap_or_else(|_| "https://api.moonshot.cn".to_string());
215        let mut cfg = ProviderConfig::openai_compatible(
216            &base_url,
217            "MOONSHOT_API_KEY",
218            "moonshot-v1-8k", // Default chat model
219            None,             // No multimodal support yet
220        );
221        cfg.chat_endpoint = "/v1/chat/completions".to_string();
222        cfg
223    }
224
225    /// Anthropic Claude configuration - requires special handling
226    pub fn anthropic() -> ProviderConfig {
227        use std::collections::HashMap;
228
229        let mut headers = HashMap::new();
230        headers.insert("Content-Type".to_string(), "application/json".to_string());
231        headers.insert("anthropic-version".to_string(), "2023-06-01".to_string());
232        // Note: Anthropic uses x-api-key instead of Authorization: Bearer
233
234        let mut role_mapping = HashMap::new();
235        role_mapping.insert("System".to_string(), "system".to_string());
236        role_mapping.insert("User".to_string(), "user".to_string());
237        role_mapping.insert("Assistant".to_string(), "assistant".to_string());
238
239        ProviderConfig {
240            base_url: "https://api.anthropic.com/v1".to_string(),
241            api_key_env: "ANTHROPIC_API_KEY".to_string(),
242            chat_endpoint: "/messages".to_string(),
243            chat_model: "claude-3-5-sonnet-20241022".to_string(), // Default chat model
244            multimodal_model: Some("claude-3-5-sonnet-20241022".to_string()), // Multimodal support
245            upload_endpoint: None,
246            upload_size_limit: None,
247            models_endpoint: None, // Claude doesn't have a public model list endpoint
248            headers,
249            field_mapping: crate::provider::config::FieldMapping {
250                messages_field: "messages".to_string(),
251                model_field: "model".to_string(),
252                role_mapping,
253                response_content_path: "content.0.text".to_string(), // Claude's response format is different
254            },
255        }
256    }
257
258    /// OpenRouter configuration
259    /// 
260    /// OpenRouter is a unified gateway for multiple AI models with OpenAI-compatible API.
261    /// Base URL: https://openrouter.ai/api/v1
262    /// Documentation: https://openrouter.ai/docs/api-reference/overview
263    pub fn openrouter() -> ProviderConfig {
264        ProviderConfig::openai_compatible(
265            "https://openrouter.ai/api/v1",
266            "OPENROUTER_API_KEY",
267            "openai/gpt-3.5-turbo", // Default model
268            Some("openai/gpt-4o"), // Multimodal model
269        )
270    }
271
272    /// Replicate configuration
273    /// 
274    /// Replicate provides access to various AI models with OpenAI-compatible API.
275    /// Base URL: https://api.replicate.com/v1
276    /// Documentation: https://replicate.com/docs/reference/http
277    pub fn replicate() -> ProviderConfig {
278        ProviderConfig::openai_compatible(
279            "https://api.replicate.com/v1",
280            "REPLICATE_API_TOKEN",
281            "meta/llama-2-7b-chat", // Default model
282            Some("meta/llama-2-7b-chat"), // Multimodal model
283        )
284    }
285
286    /// 智谱AI (GLM) configuration
287    /// 
288    /// 智谱AI provides GLM series models with OpenAI-compatible API.
289    /// Base URL: https://open.bigmodel.cn/api/paas/v4
290    /// Documentation: https://docs.bigmodel.cn/cn/api/introduction
291    pub fn zhipu_ai() -> ProviderConfig {
292        ProviderConfig::openai_compatible(
293            "https://open.bigmodel.cn/api/paas/v4",
294            "ZHIPU_API_KEY",
295            "glm-4", // Default model
296            Some("glm-4v"), // Multimodal model
297        )
298    }
299
300    /// MiniMax configuration
301    /// 
302    /// MiniMax provides AI models with OpenAI-compatible API.
303    /// Base URL: https://api.minimax.chat/v1
304    /// Documentation: https://www.minimax.io/platform/document/ChatCompletion
305    pub fn minimax() -> ProviderConfig {
306        ProviderConfig::openai_compatible(
307            "https://api.minimax.chat/v1",
308            "MINIMAX_API_KEY",
309            "abab6.5-chat", // Default model
310            Some("abab6.5-chat"), // Multimodal model
311        )
312    }
313}
ai_lib/provider/configs.rs

ai_lib/provider/
configs.rs
