From 57e3ee4ceface0de202722140aed6f0877d0e103 Mon Sep 17 00:00:00 2001
From: chiyi <cyi.cc@outlook.com>
Date: Thu, 9 Jul 2026 16:26:18 +0800
Subject: [PATCH] feat(grok): route only the conversations/new submit through
 the proxy

Reference-frame upload, the mp4 download and OpenAsset (/content) streaming now egress on the local IP; only the generate submit uses proxy.url. Verified live: submit egress 98.97.26.37 (proxy) vs upload/download 76.209.9.65 (local), 3.47MB video generated OK.
---
 backend/internal/provider/grok/client.go | 11 +++++++++--
 backend/internal/provider/grok/video.go  | 19 +++++++++++++------
 2 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/backend/internal/provider/grok/client.go b/backend/internal/provider/grok/client.go
index ad2fe06..fef7954 100644
--- a/backend/internal/provider/grok/client.go
+++ b/backend/internal/provider/grok/client.go
@@ -689,7 +689,14 @@ func (c *Client) applyHeaders(req *http.Request, token string, extra map[string]
 	req.Header = h
 }
 
-func (c *Client) newTLSClient() (tlsclient.HttpClient, error) {
+func (c *Client) newTLSClient() (tlsclient.HttpClient, error) { return c.newTLSClientP(true) }
+
+// newDirectTLSClient egresses on the local IP (never the proxy). Used for
+// reference-frame upload and result (video) download; only the generate submit
+// (/rest/app-chat/conversations/new) uses the proxy.
+func (c *Client) newDirectTLSClient() (tlsclient.HttpClient, error) { return c.newTLSClientP(false) }
+
+func (c *Client) newTLSClientP(useProxy bool) (tlsclient.HttpClient, error) {
 	options := []tlsclient.HttpClientOption{
 		// Video generation streams inline until progress=100; a 15s clip can take
 		// several minutes, so allow up to 10m (caller's genCtx caps at 12m).
@@ -697,7 +704,7 @@ func (c *Client) newTLSClient() (tlsclient.HttpClient, error) {
 		tlsclient.WithClientProfile(profiles.Chrome_133),
 		tlsclient.WithRandomTLSExtensionOrder(),
 	}
-	if c.proxy != "" {
+	if useProxy && c.proxy != "" {
 		options = append(options, tlsclient.WithProxyUrl(c.proxy))
 	}
 	return tlsclient.NewHttpClient(tlsclient.NewNoopLogger(), options...)
diff --git a/backend/internal/provider/grok/video.go b/backend/internal/provider/grok/video.go
index 6fbc4ae..1648aa8 100644
--- a/backend/internal/provider/grok/video.go
+++ b/backend/internal/provider/grok/video.go
@@ -54,7 +54,13 @@ func (c *Client) GenerateVideo(ctx context.Context, token, prompt, aspectRatio,
 		seconds = 10
 	}
 
-	client, err := c.newTLSClient()
+	// Only the conversations/new submit egresses via the proxy; reference-frame
+	// upload and the mp4 download run on the local IP.
+	submitClient, err := c.newTLSClient()
+	if err != nil {
+		return nil, nil, err
+	}
+	directClient, err := c.newDirectTLSClient()
 	if err != nil {
 		return nil, nil, err
 	}
@@ -65,7 +71,7 @@ func (c *Client) GenerateVideo(ctx context.Context, token, prompt, aspectRatio,
 		if len(f) == 0 {
 			continue
 		}
-		url, upErr := c.uploadImage(ctx, client, token, f)
+		url, upErr := c.uploadImage(ctx, directClient, token, f)
 		if upErr != nil {
 			return nil, nil, upErr
 		}
@@ -87,7 +93,7 @@ func (c *Client) GenerateVideo(ctx context.Context, token, prompt, aspectRatio,
 		if ctx.Err() != nil {
 			return nil, nil, ctx.Err()
 		}
-		pid, cpErr := c.createPost(ctx, client, token, prompt)
+		pid, cpErr := c.createPost(ctx, submitClient, token, prompt)
 		if cpErr != nil {
 			lastErr = cpErr
 			continue
@@ -116,7 +122,7 @@ func (c *Client) GenerateVideo(ctx context.Context, token, prompt, aspectRatio,
 			},
 		}
 
-		body, psErr := c.postStream(ctx, client, token, "/rest/app-chat/conversations/new", payload)
+		body, psErr := c.postStream(ctx, submitClient, token, "/rest/app-chat/conversations/new", payload)
 		if psErr != nil {
 			// Transient HTTP/2 stream resets etc. — retry.
 			lastErr = psErr
@@ -169,7 +175,7 @@ func (c *Client) GenerateVideo(ctx context.Context, token, prompt, aspectRatio,
 	if !downloadResult {
 		return nil, meta, nil
 	}
-	data, err := c.download(ctx, client, token, fullURL)
+	data, err := c.download(ctx, directClient, token, fullURL)
 	if err != nil {
 		return nil, nil, err
 	}
@@ -342,7 +348,8 @@ func (c *Client) OpenAsset(ctx context.Context, token, url string) (io.ReadClose
 	if token == "" {
 		return nil, "", ErrAuth
 	}
-	client, err := c.newTLSClient()
+	// Asset streaming is pure bandwidth (no anti-bot) — egress on the local IP.
+	client, err := c.newDirectTLSClient()
 	if err != nil {
 		return nil, "", err
 	}