Initial commit: Luminary AI Gateway

OpenAI-compatible AI gateway with Vue admin UI, multi-provider egress,
ingress key governance, monitoring, and security controls.

Co-authored-by: Cursor <cursoragent@cursor.com>

internal/handler/replicate/handler.go

@@ -0,0 +1,159 @@
+package replicate
+
+import (
+	"bufio"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/rose_cat707/luminary/internal/model"
+	"github.com/rose_cat707/luminary/internal/prediction"
+)
+
+type Handler struct {
+	svc     *prediction.Service
+	baseURL func() string
+}
+
+func New(svc *prediction.Service, baseURL func() string) *Handler {
+	return &Handler{svc: svc, baseURL: baseURL}
+}
+
+func (h *Handler) Register(r *gin.RouterGroup) {
+	r.POST("/predictions", h.Create)
+	r.GET("/predictions", h.List)
+	r.GET("/predictions/:id", h.Get)
+	r.POST("/predictions/:id/cancel", h.Cancel)
+	r.GET("/predictions/:id/stream", h.Stream)
+}
+
+func (h *Handler) ingress(c *gin.Context) (*model.IngressKey, bool) {
+	v, ok := c.Get("ingress_key")
+	if !ok {
+		return nil, false
+	}
+	k, ok := v.(*model.IngressKey)
+	return k, ok
+}
+
+func (h *Handler) Create(c *gin.Context) {
+	ingress, ok := h.ingress(c)
+	if !ok {
+		c.JSON(http.StatusUnauthorized, gin.H{"detail": "unauthorized"})
+		return
+	}
+	var req prediction.CreateRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"detail": "invalid request body"})
+		return
+	}
+	wait := strings.Contains(c.GetHeader("Prefer"), "wait")
+	p, err := h.svc.Create(c.Request.Context(), ingress, req, c.ClientIP(), wait)
+	if err != nil {
+		code := http.StatusBadRequest
+		if strings.Contains(err.Error(), "not found") {
+			code = http.StatusNotFound
+		}
+		c.JSON(code, gin.H{"detail": err.Error()})
+		return
+	}
+	c.JSON(http.StatusCreated, h.svc.APIView(p, h.baseURL()))
+}
+
+func (h *Handler) Get(c *gin.Context) {
+	ingress, ok := h.ingress(c)
+	if !ok {
+		c.JSON(http.StatusUnauthorized, gin.H{"detail": "unauthorized"})
+		return
+	}
+	p, err := h.svc.Get(c.Request.Context(), c.Param("id"), ingress)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"detail": err.Error()})
+		return
+	}
+	c.JSON(http.StatusOK, h.svc.APIView(p, h.baseURL()))
+}
+
+func (h *Handler) List(c *gin.Context) {
+	ingress, ok := h.ingress(c)
+	if !ok {
+		c.JSON(http.StatusUnauthorized, gin.H{"detail": "unauthorized"})
+		return
+	}
+	items, err := h.svc.List(c.Request.Context(), ingress, c.Query("cursor"), 100)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"detail": err.Error()})
+		return
+	}
+	results := make([]map[string]any, 0, len(items))
+	for _, p := range items {
+		results = append(results, h.svc.APIView(p, h.baseURL()))
+	}
+	c.JSON(http.StatusOK, gin.H{"results": results})
+}
+
+func (h *Handler) Cancel(c *gin.Context) {
+	ingress, ok := h.ingress(c)
+	if !ok {
+		c.JSON(http.StatusUnauthorized, gin.H{"detail": "unauthorized"})
+		return
+	}
+	p, err := h.svc.Cancel(c.Request.Context(), c.Param("id"), ingress)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"detail": err.Error()})
+		return
+	}
+	c.JSON(http.StatusOK, h.svc.APIView(p, h.baseURL()))
+}
+
+func (h *Handler) Stream(c *gin.Context) {
+	ingress, ok := h.ingress(c)
+	if !ok {
+		c.JSON(http.StatusUnauthorized, gin.H{"detail": "unauthorized"})
+		return
+	}
+	id := c.Param("id")
+	if _, err := h.svc.Get(c.Request.Context(), id, ingress); err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"detail": err.Error()})
+		return
+	}
+
+	c.Header("Content-Type", "text/event-stream")
+	c.Header("Cache-Control", "no-cache")
+	c.Header("Connection", "keep-alive")
+	c.Status(http.StatusOK)
+
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		return
+	}
+	w := bufio.NewWriter(c.Writer)
+	ch := h.svc.Hub().Subscribe(id)
+	defer h.svc.Hub().Unsubscribe(id, ch)
+
+	for {
+		select {
+		case <-c.Request.Context().Done():
+			return
+		case ev, open := <-ch:
+			if !open {
+				return
+			}
+			payload, _ := json.Marshal(map[string]any{
+				"id": id, "status": ev.Status, "logs": ev.Logs,
+				"metrics": ev.Metrics, "output": ev.Output, "error": ev.Error,
+			})
+			_, _ = fmt.Fprintf(w, "data: %s\n\n", payload)
+			w.Flush()
+			flusher.Flush()
+			if ev.Done {
+				_, _ = fmt.Fprintf(w, "event: done\ndata: {}\n\n")
+				w.Flush()
+				flusher.Flush()
+				return
+			}
+		}
+	}
+}