Merge pull request #3232 from Ajinkya-Ghuge/fix/deepseek-model-routing

rust/crates/api/src/providers/mod.rs

@@ -296,6 +296,20 @@ pub fn metadata_for_model(model: &str) -> Option<ProviderMetadata> {
     None
 }
 
+
+
+
+#[must_use]
+pub fn strip_provider_prefix(canonical_model: &str) -> String {
+    if let Some(pos) = canonical_model.find('/') {
+        canonical_model[pos + 1..].to_string()
+    } else {
+        canonical_model.to_string()
+    }
+}
+
+
+
 #[must_use]
 pub fn provider_diagnostics_for_model(model: &str) -> ProviderDiagnostics {
     let resolved_model = resolve_model_alias(model);

rust/crates/api/src/providers/openai_compat.rs

@@ -16,7 +16,8 @@ use crate::types::{
     ToolChoice, ToolDefinition, ToolResultContentBlock, Usage,
 };
 
-use super::{preflight_message_request, Provider, ProviderFuture};
+use super::{preflight_message_request, Provider, ProviderFuture, resolve_model_alias, strip_provider_prefix};
+
 
 pub const DEFAULT_XAI_BASE_URL: &str = "https://api.x.ai/v1";
 pub const DEFAULT_OPENAI_BASE_URL: &str = "https://api.openai.com/v1";
@@ -214,15 +215,74 @@ impl OpenAiCompatClient {
     pub async fn send_message(
     &self,
     request: &MessageRequest,
-    ) -> Result<MessageResponse, ApiError> {
+) -> Result<MessageResponse, ApiError> {
-        let request = MessageRequest {
+    // 1. Keep track of what Claw originally asked for
+    let original_model = request.model.clone();
+    let canonical = resolve_model_alias(&request.model);
+    
+    // 2. Clean the model string (e.g., "openai/deepseek-v4-flash" -> "deepseek-v4-flash")
+    let downstream_model = strip_provider_prefix(&canonical);
+
+    let mut request = MessageRequest {
         stream: false,
         ..request.clone()
     };
+    request.model = downstream_model; // Use the clean name for the API payload
+    
     preflight_message_request(&request)?;
     let response = self.send_with_retry(&request).await?;
     let request_id = request_id_from_headers(response.headers());
     let body = response.text().await.map_err(ApiError::from)?;
+
+    // Some backends return {"error":{"message":"...","type":"...","code":...}}
+    // instead of a valid completion object. Check for this before attempting
+    // full deserialization so the user sees the actual error, not a cryptic.
+    if let Ok(raw) = serde_json::from_str::<serde_json::Value>(&body) {
+        if let Some(err_obj) = raw.get("error") {
+            let msg = err_obj
+                .get("message")
+                .and_then(|m| m.as_str())
+                .unwrap_or("provider returned an error")
+                .to_string();
+            let code = err_obj
+                .get("code")
+                .and_then(serde_json::Value::as_u64)
+                .map(|c| c as u16);
+            return Err(ApiError::Api {
+                status: reqwest::StatusCode::from_u16(code.unwrap_or(400))
+                    .unwrap_or(reqwest::StatusCode::BAD_REQUEST),
+                error_type: err_obj
+                    .get("type")
+                    .and_then(|t| t.as_str())
+                    .map(str::to_owned),
+                message: Some(msg),
+                request_id,
+                body,
+                retryable: false,
+                suggested_action: suggested_action_for_status(
+                    reqwest::StatusCode::from_u16(code.unwrap_or(400))
+                        .unwrap_or(reqwest::StatusCode::BAD_REQUEST),
+                ),
+                retry_after: None,
+            });
+        }
+    }
+
+    // Pass original_model to the deserializer error context so debugging logs are accurate
+    let payload = serde_json::from_str::<ChatCompletionResponse>(&body).map_err(|error| {
+        ApiError::json_deserialize(self.config.provider_name, &original_model, &body, error)
+    })?;
+
+    let mut normalized = normalize_response(&request.model, payload)?;
+    if normalized.request_id.is_none() {
+        normalized.request_id = request_id;
+    }
+
+    // 3. CRITICAL: Put the original model string back so Claw's internal routing stays happy
+    normalized.model = original_model; 
+
+    Ok(normalized)
+}
         // Some backends return {"error":{"message":"...","type":"...","code":...}}
         // instead of a valid completion object. Check for this before attempting
         // full deserialization so the user sees the actual error, not a cryptic
@@ -267,23 +327,32 @@ impl OpenAiCompatClient {
         Ok(normalized)
     }
 
-    pub async fn stream_message(
+pub async fn stream_message(
     &self,
     request: &MessageRequest,
-    ) -> Result<MessageStream, ApiError> {
+) -> Result<MessageStream, ApiError> {
-        preflight_message_request(request)?;
+    // 1. Keep track of the original model name
-        let response = self
+    let original_model = request.model.clone();
-            .send_with_retry(&request.clone().with_streaming())
+    let canonical = resolve_model_alias(&request.model);
-            .await?;
+    
+    // 2. Clean it up for DeepSeek
+    let downstream_model = strip_provider_prefix(&canonical);
+
+    let mut streaming_request = request.clone().with_streaming();
+    streaming_request.model = downstream_model;
+
+    preflight_message_request(&streaming_request)?;
+    let response = self.send_with_retry(&streaming_request).await?;
+
     Ok(MessageStream {
         request_id: request_id_from_headers(response.headers()),
         response,
-            parser: OpenAiSseParser::with_context(self.config.provider_name, request.model.clone()),
+        parser: OpenAiSseParser::with_context(self.config.provider_name, original_model.clone()),
         pending: VecDeque::new(),
         done: false,
-            state: StreamState::new(request.model.clone()),
+        state: StreamState::new(original_model), // 3. Use the original name here
     })
-    }
+}
 
     async fn send_with_retry(
         &self,