feat(providers): add NEAR AI Cloud provider (#9352)

Signed-off-by: Pierre LE GUEN <26087574+PierreLeGuen@users.noreply.github.com>
2026-05-22 19:56:31 +02:00
parent ee65a39d68
commit c434c84d23
5 changed files with 180 additions and 7 deletions
@@ -834,6 +834,28 @@ mod tests {
        assert_eq!(config.models[0].context_limit, 131072);
    }

+    #[test]
+    fn test_nearai_json_deserializes() {
+        let json = include_str!("../providers/declarative/nearai.json");
+        let config: DeclarativeProviderConfig =
+            serde_json::from_str(json).expect("nearai.json should parse");
+        assert_eq!(config.name, "nearai");
+        assert_eq!(config.display_name, "NEAR AI Cloud");
+        assert!(matches!(config.engine, ProviderEngine::OpenAI));
+        assert_eq!(config.api_key_env, "NEARAI_API_KEY");
+        assert_eq!(config.base_url, "https://cloud-api.near.ai/v1");
+        assert_eq!(config.catalog_provider_id, Some("nearai".to_string()));
+        assert_eq!(config.dynamic_models, Some(true));
+        assert_eq!(config.supports_streaming, Some(true));
+        assert!(config.preserves_thinking);
+        assert_eq!(
+            config.model_doc_link,
+            Some("https://docs.near.ai/".to_string())
+        );
+        assert_eq!(config.models[0].name, "zai-org/GLM-5.1-FP8");
+        assert!(config.models[0].reasoning);
+    }
+
    #[test]
    fn test_vercel_ai_gateway_json_deserializes() {
        let json = include_str!("../providers/declarative/vercel_ai_gateway.json");
@@ -0,0 +1,43 @@
+{
+  "name": "nearai",
+  "engine": "openai",
+  "display_name": "NEAR AI Cloud",
+  "description": "TEE-backed private inference through NEAR AI Cloud's OpenAI-compatible API.",
+  "api_key_env": "NEARAI_API_KEY",
+  "base_url": "https://cloud-api.near.ai/v1",
+  "catalog_provider_id": "nearai",
+  "dynamic_models": true,
+  "models": [
+    {
+      "name": "zai-org/GLM-5.1-FP8",
+      "context_limit": 202752,
+      "reasoning": true
+    },
+    {
+      "name": "Qwen/Qwen3.6-35B-A3B-FP8",
+      "context_limit": 262144,
+      "reasoning": true
+    },
+    {
+      "name": "Qwen/Qwen3.5-122B-A10B",
+      "context_limit": 131072,
+      "reasoning": true
+    },
+    {
+      "name": "Qwen/Qwen3-VL-30B-A3B-Instruct",
+      "context_limit": 256000
+    },
+    {
+      "name": "google/gemma-4-31B-it",
+      "context_limit": 262144
+    }
+  ],
+  "preserves_thinking": true,
+  "supports_streaming": true,
+  "model_doc_link": "https://docs.near.ai/",
+  "setup_steps": [
+    "Create or sign in to your NEAR AI Cloud account at https://cloud.near.ai",
+    "Create an API key",
+    "Copy the key and paste it above"
+  ]
+}
@@ -284,6 +284,30 @@ mod tests {
        );
    }

+    #[tokio::test]
+    async fn test_nearai_declarative_provider_registry_wiring() {
+        let nearai = get_from_registry("nearai")
+            .await
+            .expect("nearai provider should be registered");
+        let meta = nearai.metadata();
+
+        assert_eq!(nearai.provider_type(), ProviderType::Declarative);
+        assert!(nearai.supports_inventory_refresh());
+        assert_eq!(meta.display_name, "NEAR AI Cloud");
+        assert_eq!(meta.default_model, "zai-org/GLM-5.1-FP8");
+        assert_eq!(meta.model_doc_link, "https://docs.near.ai/");
+        assert!(!meta.setup_steps.is_empty());
+
+        let api_key = meta
+            .config_keys
+            .iter()
+            .find(|k| k.name == "NEARAI_API_KEY")
+            .expect("NEARAI_API_KEY config key should exist");
+        assert!(api_key.required, "NEARAI_API_KEY should be required");
+        assert!(api_key.secret, "NEARAI_API_KEY should be secret");
+        assert!(api_key.primary, "NEARAI_API_KEY should be primary");
+    }
+
    #[tokio::test]
    async fn test_openai_compatible_providers_config_keys() {
        let providers_list = providers().await;
@@ -511,23 +511,51 @@ impl OpenAiProvider {
        "lmstudio",
        "mistral",
        "moonshot",
+        "nearai",
        "ovhcloud",
    ];

-    fn sanitize_request_for_compat(&self, mut payload: serde_json::Value) -> serde_json::Value {
-        if !Self::PROVIDERS_NEEDING_MAX_TOKENS_REMAP.contains(&self.name.as_str()) {
-            return payload;
-        }
+    const PROVIDERS_NEEDING_STANDARD_CHAT_PARAMS: &[&str] = &["nearai"];

+    fn sanitize_request_for_compat(&self, mut payload: serde_json::Value) -> serde_json::Value {
        if let Some(obj) = payload.as_object_mut() {
-            if let Some(value) = obj.remove("max_completion_tokens") {
-                obj.entry("max_tokens").or_insert(value);
+            if Self::PROVIDERS_NEEDING_MAX_TOKENS_REMAP.contains(&self.name.as_str()) {
+                if let Some(value) = obj.remove("max_completion_tokens") {
+                    obj.entry("max_tokens").or_insert(value);
+                }
+            }
+
+            if Self::PROVIDERS_NEEDING_STANDARD_CHAT_PARAMS.contains(&self.name.as_str()) {
+                let model_name = obj.get("model").and_then(|model| model.as_str());
+                if !model_name.is_some_and(Self::is_responses_model) {
+                    obj.remove("reasoning_effort");
+                }
+
+                if let Some(messages) = obj.get_mut("messages").and_then(|m| m.as_array_mut()) {
+                    for message in messages {
+                        if message
+                            .get("role")
+                            .and_then(|role| role.as_str())
+                            .is_some_and(|role| role == "developer")
+                        {
+                            message["role"] = serde_json::Value::String("system".to_string());
+                        }
+                    }
+                }
            }
        }

        payload
    }

+    fn should_use_responses_api_for_provider(&self, model_name: &str) -> bool {
+        if Self::PROVIDERS_NEEDING_STANDARD_CHAT_PARAMS.contains(&self.name.as_str()) {
+            return false;
+        }
+
+        Self::should_use_responses_api(model_name, &self.base_path)
+    }
+
    fn map_base_path(base_path: &str, target: &str, fallback: &str) -> String {
        let normalized = Self::normalize_base_path(base_path);
        if normalized.ends_with(target) || normalized.contains(&format!("/{target}")) {
@@ -748,7 +776,7 @@ impl Provider for OpenAiProvider {
        messages: &[Message],
        tools: &[Tool],
    ) -> Result<MessageStream, ProviderError> {
-        if Self::should_use_responses_api(&model_config.model_name, &self.base_path) {
+        if self.should_use_responses_api_for_provider(&model_config.model_name) {
            let mut payload = create_responses_request(model_config, system, messages, tools)?;
            payload["stream"] = serde_json::Value::Bool(self.supports_streaming);

@@ -1029,6 +1057,61 @@ mod tests {
        assert_eq!(result, payload);
    }

+    #[test]
+    fn sanitize_nearai_reasoning_chat_params() {
+        let provider = make_provider("nearai");
+        let payload = json!({
+            "model": "Qwen/Qwen3.6-35B-A3B-FP8",
+            "messages": [
+                {
+                    "role": "developer",
+                    "content": "system instructions"
+                },
+                {
+                    "role": "user",
+                    "content": "hello"
+                }
+            ],
+            "reasoning_effort": "medium",
+            "max_completion_tokens": 16384
+        });
+
+        let result = provider.sanitize_request_for_compat(payload);
+        let obj = result.as_object().unwrap();
+
+        assert!(!obj.contains_key("reasoning_effort"));
+        assert!(!obj.contains_key("max_completion_tokens"));
+        assert_eq!(obj.get("max_tokens").unwrap(), &json!(16384));
+        assert_eq!(obj["messages"][0]["role"], "system");
+        assert_eq!(obj["messages"][1]["role"], "user");
+    }
+
+    #[test]
+    fn sanitize_nearai_preserves_openai_reasoning_effort() {
+        let provider = make_provider("nearai");
+        let payload = json!({
+            "model": "openai/gpt-5",
+            "messages": [],
+            "reasoning_effort": "medium",
+            "max_completion_tokens": 16384
+        });
+
+        let result = provider.sanitize_request_for_compat(payload);
+        let obj = result.as_object().unwrap();
+
+        assert_eq!(obj.get("reasoning_effort"), Some(&json!("medium")));
+        assert!(!obj.contains_key("max_completion_tokens"));
+        assert_eq!(obj.get("max_tokens").unwrap(), &json!(16384));
+    }
+
+    #[test]
+    fn nearai_uses_chat_completions_for_openai_reasoning_models() {
+        let provider = make_provider("nearai");
+
+        assert!(!provider.should_use_responses_api_for_provider("openai/gpt-5"));
+        assert!(!provider.should_use_responses_api_for_provider("openai/o3"));
+    }
+
    #[test]
    fn responses_api_routing_uses_model_family_unless_path_forces_chat() {
        for (model_name, base_path, expected) in [
@@ -39,6 +39,7 @@ goose is compatible with a wide range of LLM providers, allowing you to choose a
 | [LiteLLM](https://docs.litellm.ai/docs/) | LiteLLM proxy supporting multiple models with automatic prompt caching and unified API access. | `LITELLM_HOST`, `LITELLM_BASE_PATH` (optional), `LITELLM_API_KEY` (optional), `LITELLM_CUSTOM_HEADERS` (optional), `LITELLM_TIMEOUT` (optional) |
 | [LM Studio](https://lmstudio.ai/)                                          | Run local models with LM Studio's OpenAI-compatible server. **Because this provider runs locally, you must first [download a model](#local-llms).**                                                           | None required. Connects to local server at `localhost:1234` by default.                                                                                                             |
 | [Mistral AI](https://mistral.ai/)                                           | Provides access to Mistral models including general-purpose models, specialized coding models (Codestral), and multimodal models (Pixtral).                                                                   | `MISTRAL_API_KEY`                                                                                                 |
+| [NEAR AI Cloud](https://cloud.near.ai/)                                     | TEE-backed private inference through an OpenAI-compatible API with dynamic model discovery.                                                                                                                   | `NEARAI_API_KEY`                                                                                                                                                                  |
 | [Novita AI](https://novita.ai/)                                             | 90+ open-source models with OpenAI-compatible API and competitive pricing. Supports Kimi K2.5, DeepSeek, GLM, MiniMax, Qwen, and more.                                                                       | `NOVITA_API_KEY`                                                                                                  |
 | [Ollama](https://ollama.com/)                                               | Local model runner supporting Qwen, Llama, DeepSeek, and other open-source models. **Because this provider runs locally, you must first [download and run a model](#local-llms).**  | `OLLAMA_HOST`                                                                                                                                                                       |
 | [Ollama Cloud](https://ollama.com/)                                         | Access hosted models on ollama.com via OpenAI-compatible API. Requires an Ollama account and API key.  | `OLLAMA_CLOUD_API_KEY`                                                                                                                                                                       |