Fix OpenAIProxy handler

pkg/handlers.go

@@ -1,8 +1,10 @@
 package llamactl
 
 import (
+	"bytes"
 	"encoding/json"
 	"fmt"
+	"io"
 	"net/http"
 	"os/exec"
 	"strconv"
@@ -456,9 +458,17 @@ func (h *Handler) ProxyToInstance() http.HandlerFunc {
 // OpenAIProxy godoc
 func (h *Handler) OpenAIProxy() http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
-		// Extract model name from request body
+		// Read the entire body first
+		bodyBytes, err := io.ReadAll(r.Body)
+		if err != nil {
+			http.Error(w, "Failed to read request body", http.StatusBadRequest)
+			return
+		}
+		r.Body.Close()
+
+		// Parse the body to extract model name
 		var requestBody map[string]any
-		if err := json.NewDecoder(r.Body).Decode(&requestBody); err != nil {
+		if err := json.Unmarshal(bodyBytes, &requestBody); err != nil {
 			http.Error(w, "Invalid request body", http.StatusBadRequest)
 			return
 		}
@@ -487,6 +497,10 @@ func (h *Handler) OpenAIProxy() http.HandlerFunc {
 			return
 		}
 
+		// Recreate the request body from the bytes we read
+		r.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+		r.ContentLength = int64(len(bodyBytes))
+
 		proxy.ServeHTTP(w, r)
 	}
 }

pkg/routes.go

@@ -50,13 +50,15 @@ func SetupRouter(handler *Handler) *chi.Mux {
 		})
 	})
 
-	// OpenAI-compatible endpoints (model name in request body determines routing)
-	r.Post("/v1/", handler.OpenAIProxy()) // Proxy to OpenAI-compatible endpoints based on instance name in request body
-	// r.Post("/v1/completions", handler.OpenAICompletions())
-	// r.Post("/v1/chat/completions", handler.OpenAIChatCompletions())
-	// r.Post("/v1/embeddings", handler.OpenAIEmbeddings())
-	// r.Post("/v1/rerank", handler.OpenAIRerank())
-	// r.Post("/v1/reranking", handler.OpenAIReranking())
+	// OpenAI-compatible proxy endpoint
+	// Handles all POST requests to /v1/*, including:
+	//   - /v1/completions
+	//   - /v1/chat/completions
+	//   - /v1/embeddings
+	//   - /v1/rerank
+	//   - /v1/reranking
+	// The instance/model to use is determined by the request body.
+	r.Post("/v1/*", handler.OpenAIProxy())
 
 	// Serve WebUI files
 	if err := webui.SetupWebUI(r); err != nil {