synaps_cli/runtime/openai/
registry.rs

1//! Provider registry — catalog of known OpenAI-compatible endpoints.
2//!
3//! Ported from `openai-runtime::registry`, extended to accept a config map
4//! override for API keys (checked before env vars).
5
6use serde::Deserialize;
7use super::types::ProviderConfig;
8use std::collections::BTreeMap;
9
10#[derive(Debug)]
11pub struct ProviderSpec {
12    pub key: &'static str,
13    pub name: &'static str,
14    pub base_url: &'static str,
15    pub env_vars: &'static [&'static str],
16    pub default_model: &'static str,
17    pub models: &'static [(&'static str, &'static str, &'static str)], // (id, label, tier)
18}
19
20#[derive(Debug, Clone, PartialEq, Eq)]
21pub struct ProviderModelInfo {
22    pub id: String,
23    pub name: Option<String>,
24}
25
26#[derive(Debug, Deserialize)]
27struct ProviderModelsResponse {
28    data: Vec<ProviderModelsItem>,
29}
30
31#[derive(Debug, Deserialize)]
32struct ProviderModelsItem {
33    id: String,
34    #[serde(default)]
35    name: Option<String>,
36}
37
38pub fn parse_provider_models_response(body: &str) -> Result<Vec<ProviderModelInfo>, serde_json::Error> {
39    let response: ProviderModelsResponse = serde_json::from_str(body)?;
40    Ok(response
41        .data
42        .into_iter()
43        .filter(|item| !item.id.trim().is_empty())
44        .map(|item| ProviderModelInfo {
45            id: item.id,
46            name: item.name.filter(|name| !name.trim().is_empty()),
47        })
48        .collect())
49}
50
51pub fn providers() -> &'static [ProviderSpec] {
52    static PROVIDERS: std::sync::LazyLock<Vec<ProviderSpec>> = std::sync::LazyLock::new(|| vec![
53        ProviderSpec {
54            key: "groq",
55            name: "Groq",
56            base_url: "https://api.groq.com/openai/v1",
57            env_vars: &["GROQ_API_KEY"],
58            default_model: "llama-3.3-70b-versatile",
59            models: &[
60                ("llama-3.3-70b-versatile", "Llama 3.3 70B", "S"),
61                ("llama-3.1-8b-instant", "Llama 3.1 8B", "B"),
62                ("meta-llama/llama-4-scout-17b-16e-instruct", "Llama 4 Scout", "A"),
63                ("meta-llama/llama-4-maverick-17b-128e-instruct", "Llama 4 Maverick", "S"),
64            ],
65        },
66        ProviderSpec {
67            key: "cerebras",
68            name: "Cerebras",
69            base_url: "https://api.cerebras.ai/v1",
70            env_vars: &["CEREBRAS_API_KEY"],
71            default_model: "llama3.1-8b",
72            models: &[
73                ("qwen-3-235b-a22b-instruct-2507", "Qwen3 235B", "S+"),
74                ("llama3.1-8b", "Llama 3.1 8B", "B"),
75            ],
76        },
77        ProviderSpec {
78            key: "nvidia",
79            name: "NVIDIA NIM",
80            base_url: "https://integrate.api.nvidia.com/v1",
81            env_vars: &["NVIDIA_API_KEY"],
82            default_model: "meta/llama-3.3-70b-instruct",
83            models: &[
84                ("qwen/qwen3-coder-480b-a35b-instruct", "Qwen3 Coder 480B", "S+"),
85                ("mistralai/mistral-large-3-675b-instruct-2512", "Mistral Large 675B", "A+"),
86                ("meta/llama-3.3-70b-instruct", "Llama 3.3 70B", "A"),
87                ("meta/llama-4-maverick-17b-128e-instruct", "Llama 4 Maverick", "S"),
88                ("meta/llama-4-scout-17b-16e-instruct", "Llama 4 Scout", "A"),
89                ("nvidia/llama-3.1-nemotron-ultra-253b-v1", "Nemotron Ultra 253B", "A+"),
90                ("mistralai/devstral-2-123b-instruct-2512", "Devstral 2 123B", "S+"),
91                ("minimaxai/minimax-m2.5", "MiniMax M2.5", "S+"),
92                ("stepfun-ai/step-3.5-flash", "Step 3.5 Flash", "S+"),
93            ],
94        },
95        ProviderSpec {
96            key: "sambanova",
97            name: "SambaNova",
98            base_url: "https://api.sambanova.ai/v1",
99            env_vars: &["SAMBANOVA_API_KEY"],
100            default_model: "Meta-Llama-3.3-70B-Instruct",
101            models: &[
102                ("QwQ-32B", "QwQ 32B", "A+"),
103                ("Meta-Llama-3.3-70B-Instruct", "Llama 3.3 70B", "S"),
104                ("Meta-Llama-3.1-8B-Instruct", "Llama 3.1 8B", "B"),
105                ("DeepSeek-R1", "DeepSeek R1", "S+"),
106                ("DeepSeek-R1-Distill-Llama-70B", "R1 Distill 70B", "A"),
107                ("Qwen3-32B", "Qwen3 32B", "A"),
108            ],
109        },
110        ProviderSpec {
111            key: "openrouter",
112            name: "OpenRouter",
113            base_url: "https://openrouter.ai/api/v1",
114            env_vars: &["OPENROUTER_API_KEY"],
115            default_model: "meta-llama/llama-3.3-70b-instruct",
116            models: &[
117                ("qwen/qwen3-coder", "Qwen3 Coder", "S+"),
118                ("meta-llama/llama-3.3-70b-instruct", "Llama 3.3 70B", "S"),
119                ("deepseek/deepseek-chat-v3-0324", "DeepSeek V3", "S"),
120                ("google/gemma-3-27b-it", "Gemma 3 27B", "A"),
121                ("mistralai/mistral-small-3.1-24b-instruct", "Mistral Small 3.1", "A"),
122            ],
123        },
124        ProviderSpec {
125            key: "google",
126            name: "Google AI Studio",
127            base_url: "https://generativelanguage.googleapis.com/v1beta/openai",
128            env_vars: &["GOOGLE_API_KEY"],
129            default_model: "gemini-2.5-flash",
130            models: &[
131                ("gemini-2.5-flash", "Gemini 2.5 Flash", "A+"),
132                ("gemini-2.0-flash", "Gemini 2.0 Flash", "B+"),
133                ("gemma-3-27b-it", "Gemma 3 27B", "A"),
134            ],
135        },
136        ProviderSpec {
137            key: "deepinfra",
138            name: "DeepInfra",
139            base_url: "https://api.deepinfra.com/v1/openai",
140            env_vars: &["DEEPINFRA_API_KEY", "DEEPINFRA_TOKEN"],
141            default_model: "meta-llama/Llama-3.3-70B-Instruct",
142            models: &[
143                ("meta-llama/Llama-3.3-70B-Instruct", "Llama 3.3 70B", "S"),
144                ("Qwen/Qwen2.5-Coder-32B-Instruct", "Qwen2.5 Coder 32B", "A"),
145                ("deepseek-ai/DeepSeek-V3-0324", "DeepSeek V3", "S"),
146            ],
147        },
148        ProviderSpec {
149            key: "huggingface",
150            name: "HuggingFace",
151            base_url: "https://router.huggingface.co/v1",
152            env_vars: &["HUGGINGFACE_API_KEY", "HF_TOKEN"],
153            default_model: "meta-llama/Llama-3.3-70B-Instruct",
154            models: &[
155                ("meta-llama/Llama-3.3-70B-Instruct", "Llama 3.3 70B", "S"),
156                ("Qwen/Qwen2.5-72B-Instruct", "Qwen2.5 72B", "A"),
157            ],
158        },
159        ProviderSpec {
160            key: "fireworks",
161            name: "Fireworks AI",
162            base_url: "https://api.fireworks.ai/inference/v1",
163            env_vars: &["FIREWORKS_API_KEY"],
164            default_model: "accounts/fireworks/models/llama-v3p3-70b-instruct",
165            models: &[
166                ("accounts/fireworks/models/llama-v3p3-70b-instruct", "Llama 3.3 70B", "S"),
167                ("accounts/fireworks/models/qwen2p5-coder-32b-instruct", "Qwen2.5 Coder 32B", "A"),
168            ],
169        },
170        ProviderSpec {
171            key: "hyperbolic",
172            name: "Hyperbolic",
173            base_url: "https://api.hyperbolic.xyz/v1",
174            env_vars: &["HYPERBOLIC_API_KEY"],
175            default_model: "meta-llama/Llama-3.3-70B-Instruct",
176            models: &[
177                ("meta-llama/Llama-3.3-70B-Instruct", "Llama 3.3 70B", "S"),
178                ("Qwen/Qwen2.5-Coder-32B-Instruct", "Qwen2.5 Coder 32B", "A"),
179                ("deepseek-ai/DeepSeek-V3-0324", "DeepSeek V3", "S"),
180            ],
181        },
182        ProviderSpec {
183            key: "scaleway",
184            name: "Scaleway",
185            base_url: "https://api.scaleway.ai/v1",
186            env_vars: &["SCALEWAY_API_KEY"],
187            default_model: "llama-3.3-70b-instruct",
188            models: &[
189                ("llama-3.3-70b-instruct", "Llama 3.3 70B", "S"),
190                ("qwen3-235b-a22b", "Qwen3 235B", "S+"),
191            ],
192        },
193        ProviderSpec {
194            key: "siliconflow",
195            name: "SiliconFlow",
196            base_url: "https://api.siliconflow.cn/v1",
197            env_vars: &["SILICONFLOW_API_KEY"],
198            default_model: "Qwen/Qwen3-8B",
199            models: &[
200                ("Qwen/Qwen3-8B", "Qwen3 8B", "A-"),
201                ("deepseek-ai/DeepSeek-R1", "DeepSeek R1", "S+"),
202            ],
203        },
204        ProviderSpec {
205            key: "together",
206            name: "Together AI",
207            base_url: "https://api.together.xyz/v1",
208            env_vars: &["TOGETHER_API_KEY"],
209            default_model: "meta-llama/Llama-3.3-70B-Instruct-Turbo",
210            models: &[
211                ("meta-llama/Llama-3.3-70B-Instruct-Turbo", "Llama 3.3 70B", "S"),
212                ("Qwen/Qwen2.5-Coder-32B-Instruct", "Qwen2.5 Coder 32B", "A"),
213                ("deepseek-ai/DeepSeek-V3", "DeepSeek V3", "S"),
214            ],
215        },
216        ProviderSpec {
217            key: "chutes",
218            name: "Chutes AI",
219            base_url: "https://llm.chutes.ai/v1",
220            env_vars: &["CHUTES_API_KEY"],
221            default_model: "deepseek-ai/DeepSeek-V3-0324",
222            models: &[
223                ("deepseek-ai/DeepSeek-V3-0324", "DeepSeek V3", "S"),
224            ],
225        },
226        ProviderSpec {
227            key: "codestral",
228            name: "Codestral (Mistral)",
229            base_url: "https://api.mistral.ai/v1",
230            env_vars: &["CODESTRAL_API_KEY"],
231            default_model: "codestral-latest",
232            models: &[
233                ("codestral-latest", "Codestral", "B+"),
234            ],
235        },
236        ProviderSpec {
237            key: "perplexity",
238            name: "Perplexity",
239            base_url: "https://api.perplexity.ai",
240            env_vars: &["PERPLEXITY_API_KEY", "PPLX_API_KEY"],
241            default_model: "llama-3.1-sonar-large-128k-online",
242            models: &[
243                ("llama-3.1-sonar-large-128k-online", "Sonar Large", "A+"),
244            ],
245        },
246        ProviderSpec {
247            key: "ovhcloud",
248            name: "OVHcloud",
249            base_url: "https://oai.endpoints.kepler.ai.cloud.ovh.net/v1",
250            env_vars: &["OVH_AI_ENDPOINTS_ACCESS_TOKEN"],
251            default_model: "Meta-Llama-3.3-70B-Instruct",
252            models: &[
253                ("Meta-Llama-3.3-70B-Instruct", "Llama 3.3 70B", "S"),
254                ("Qwen/QwQ-32B", "QwQ 32B", "A+"),
255            ],
256        },
257    ]);
258    &PROVIDERS
259}
260
261/// Look up a provider by key and resolve its API key (config override first, then env vars).
262pub fn resolve_provider(
263    key: &str,
264    overrides: &BTreeMap<String, String>,
265) -> Option<(ProviderConfig, &'static str)> {
266    let specs = providers();
267    let spec = specs.into_iter().find(|s| s.key == key)?;
268    let api_key = resolve_api_key(spec.key, spec.env_vars, overrides)?;
269    Some((
270        ProviderConfig {
271            base_url: spec.base_url.to_string(),
272            api_key,
273            model: spec.default_model.to_string(),
274            provider: spec.key.to_string(),
275        },
276        spec.default_model,
277    ))
278}
279
280/// Resolve a provider + specific model.
281pub fn resolve_provider_model(
282    key: &str,
283    model: &str,
284    overrides: &BTreeMap<String, String>,
285) -> Option<ProviderConfig> {
286    // Special case: local provider — dynamic URL from config/env
287    if key == "local" {
288        return Some(resolve_local(model, overrides));
289    }
290    let specs = providers();
291    let spec = specs.into_iter().find(|s| s.key == key)?;
292    let api_key = resolve_api_key(spec.key, spec.env_vars, overrides)?;
293    Some(ProviderConfig {
294        base_url: spec.base_url.to_string(),
295        api_key,
296        model: model.to_string(),
297        provider: spec.key.to_string(),
298    })
299}
300
301/// Resolve `"provider/model"` shorthand.
302pub fn resolve_shorthand(s: &str, overrides: &BTreeMap<String, String>) -> Option<ProviderConfig> {
303    let (provider_key, model) = s.split_once('/')?;
304    resolve_provider_model(provider_key, model, overrides)
305}
306
307/// Resolve `"openai-codex/model"` shorthand if Codex OAuth is configured.
308pub fn resolve_codex_shorthand(s: &str) -> Option<ProviderConfig> {
309    let (provider_key, model) = s.split_once('/')?;
310    if provider_key != "openai-codex" {
311        return None;
312    }
313    let token = std::env::var("OPENAI_CODEX_ACCESS_TOKEN")
314        .ok()
315        .filter(|v| !v.is_empty());
316    Some(ProviderConfig {
317        base_url: "https://chatgpt.com/backend-api".to_string(),
318        api_key: token.unwrap_or_default(),
319        model: model.to_string(),
320        provider: "openai-codex".to_string(),
321    })
322}
323
324/// Resolve a local model endpoint (Ollama, LM Studio, vLLM, llama.cpp, etc.)
325///
326/// URL resolution: `provider.local.url` in config → `LOCAL_ENDPOINT` env → `http://localhost:11434/v1`
327/// API key: `provider.local` in config → `LOCAL_API_KEY` env → `"local"` (most local servers don't need one)
328fn resolve_local(model: &str, overrides: &BTreeMap<String, String>) -> ProviderConfig {
329    let base_url = overrides
330        .get("local.url")
331        .filter(|s| !s.is_empty())
332        .cloned()
333        .or_else(|| std::env::var("LOCAL_ENDPOINT").ok().filter(|s| !s.is_empty()))
334        .unwrap_or_else(|| "http://localhost:11434/v1".to_string());
335
336    let api_key = overrides
337        .get("local")
338        .filter(|s| !s.is_empty())
339        .cloned()
340        .or_else(|| std::env::var("LOCAL_API_KEY").ok().filter(|s| !s.is_empty()))
341        .unwrap_or_else(|| "local".to_string());
342
343    ProviderConfig {
344        base_url,
345        api_key,
346        model: model.to_string(),
347        provider: "local".to_string(),
348    }
349}
350
351pub async fn fetch_provider_models(
352    client: &reqwest::Client,
353    provider_key: &str,
354    overrides: &BTreeMap<String, String>,
355) -> Result<Vec<ProviderModelInfo>, String> {
356    let spec = providers()
357        .iter()
358        .find(|spec| spec.key == provider_key)
359        .ok_or_else(|| format!("unknown provider: {provider_key}"))?;
360    let api_key = resolve_api_key(spec.key, spec.env_vars, overrides)
361        .ok_or_else(|| format!("{} is not configured", spec.name))?;
362    let url = format!("{}/models", spec.base_url.trim_end_matches('/'));
363    let response = client
364        .get(url)
365        .bearer_auth(api_key)
366        .send()
367        .await
368        .map_err(|e| format!("request failed: {e}"))?;
369    let status = response.status();
370    let body = response
371        .text()
372        .await
373        .map_err(|e| format!("failed to read response: {e}"))?;
374    if !status.is_success() {
375        return Err(format!("model list failed: HTTP {status}"));
376    }
377    parse_provider_models_response(&body).map_err(|e| format!("failed to parse model list: {e}"))
378}
379
380/// List all providers with key status.
381pub fn list_providers(
382    overrides: &BTreeMap<String, String>,
383) -> Vec<(&'static str, &'static str, bool, usize)> {
384    providers()
385        .into_iter()
386        .map(|s| {
387            let has_key = resolve_api_key(s.key, s.env_vars, overrides).is_some();
388            (s.key, s.name, has_key, s.models.len())
389        })
390        .collect()
391}
392
393/// List models for a provider.
394pub fn list_models(key: &str) -> Option<Vec<(&'static str, &'static str, &'static str)>> {
395    let specs = providers();
396    let spec = specs.into_iter().find(|s| s.key == key)?;
397    Some(spec.models.to_vec())
398}
399
400/// Find all providers with a resolvable API key.
401pub fn configured_providers(
402    overrides: &BTreeMap<String, String>,
403) -> Vec<(&'static str, &'static str, &'static str)> {
404    providers()
405        .into_iter()
406        .filter_map(|s| {
407            resolve_api_key(s.key, s.env_vars, overrides)
408                .map(|_| (s.key, s.name, s.default_model))
409        })
410        .collect()
411}
412
413/// Resolve an API key. Config override (keyed by provider `key`) wins over env vars.
414fn resolve_api_key(
415    provider_key: &str,
416    env_vars: &[&str],
417    overrides: &BTreeMap<String, String>,
418) -> Option<String> {
419    if let Some(v) = overrides.get(provider_key) {
420        if !v.is_empty() {
421            return Some(v.clone());
422        }
423    }
424    env_vars.iter().find_map(|var| {
425        std::env::var(var).ok().filter(|v| !v.is_empty())
426    })
427}
428
429#[cfg(test)]
430mod model_list_tests {
431    use super::*;
432
433    #[test]
434    fn parses_openrouter_models_response() {
435        let json = r#"{
436            "data": [
437                { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder" },
438                { "id": "openai/gpt-oss-120b" }
439            ]
440        }"#;
441
442        let models = parse_provider_models_response(json).expect("parse models");
443        assert_eq!(models.len(), 2);
444        assert_eq!(models[0].id, "qwen/qwen3-coder");
445        assert_eq!(models[0].name.as_deref(), Some("Qwen: Qwen3 Coder"));
446        assert_eq!(models[1].id, "openai/gpt-oss-120b");
447        assert_eq!(models[1].name, None);
448    }
449}
synaps_cli/runtime/openai/registry.rs

synaps_cli/runtime/openai/
registry.rs
