feat: improve model rate limit tracking

apps/api/internal/httpapi/handlers.go

@@ -490,7 +490,7 @@ func (s *Server) estimatePricing(w http.ResponseWriter, r *http.Request) {
 	estimate, err := s.runner.Estimate(r.Context(), kind, model, body, user)
 	if err != nil {
 		if errors.Is(err, store.ErrNoModelCandidate) {
-			writeError(w, http.StatusNotFound, "no enabled platform model matches request")
+			writeError(w, statusFromRunError(err), err.Error(), store.ModelCandidateErrorCode(err))
 			return
 		}
 		s.logger.Error("estimate pricing failed", "error", err)
@@ -510,6 +510,16 @@ func (s *Server) listRateLimitWindows(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, map[string]any{"items": items})
 }
 
+func (s *Server) listModelRateLimitStatuses(w http.ResponseWriter, r *http.Request) {
+	items, err := s.store.ListModelRateLimitStatuses(r.Context())
+	if err != nil {
+		s.logger.Error("list model rate limit statuses failed", "error", err)
+		writeError(w, http.StatusInternalServerError, "list model rate limit statuses failed")
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]any{"items": items})
+}
+
 func (s *Server) createTask(kind string, compatible bool) http.Handler {
 	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		user, ok := auth.UserFromContext(r.Context())
@@ -557,7 +567,7 @@ func (s *Server) createTask(kind string, compatible bool) http.Handler {
 				if runErr != nil {
 					status := statusFromRunError(runErr)
 					errorPayload := map[string]any{
-						"code":    clients.ErrorCode(runErr),
+						"code":    runErrorCode(runErr),
 						"message": runErr.Error(),
 						"status":  status,
 					}
@@ -581,7 +591,7 @@ func (s *Server) createTask(kind string, compatible bool) http.Handler {
 			}
 			result, runErr := s.runner.Execute(r.Context(), task, user)
 			if runErr != nil {
-				writeError(w, statusFromRunError(runErr), runErr.Error(), clients.ErrorCode(runErr))
+				writeError(w, statusFromRunError(runErr), runErr.Error(), runErrorCode(runErr))
 				return
 			}
 			writeJSON(w, http.StatusOK, result.Output)
@@ -634,10 +644,14 @@ func scopeForTaskKind(kind string) string {
 
 func statusFromRunError(err error) int {
 	switch {
+	case store.ModelCandidateErrorCode(err) == "platform_cooling_down" || store.ModelCandidateErrorCode(err) == "model_cooling_down":
+		return http.StatusTooManyRequests
 	case errors.Is(err, store.ErrNoModelCandidate):
 		return http.StatusNotFound
 	case errors.Is(err, store.ErrRateLimited):
 		return http.StatusTooManyRequests
+	case clients.ErrorCode(err) == "rate_limit":
+		return http.StatusTooManyRequests
 	case errors.Is(err, store.ErrInsufficientWalletBalance):
 		return http.StatusPaymentRequired
 	default:
@@ -645,6 +659,13 @@ func statusFromRunError(err error) int {
 	}
 }
 
+func runErrorCode(err error) string {
+	if errors.Is(err, store.ErrNoModelCandidate) {
+		return store.ModelCandidateErrorCode(err)
+	}
+	return clients.ErrorCode(err)
+}
+
 func (s *Server) listTasks(w http.ResponseWriter, r *http.Request) {
 	user, ok := auth.UserFromContext(r.Context())
 	if !ok {