feat: AI chat landing page

[derrekcoleman]

Update landing page to an AI chat
Suggest questions to help users discover what to ask
Offer follow-up questions relevant to conversation as it unfolds
Update existing "Ask AI" pop-up styling and functionality to match

Landing page:

Ask AI button pop-up:

2025-10-23.09-02-39.mp4

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
docs	Ready	Preview	Comment	Nov 10, 2025 10:34pm

[derrekcoleman]

This is where we can hard-code the questions we want the initial suggestions and follow-up suggestions to draw from.

[andrewxhill]

it's looking great!

minor suggestion on the interface itself. bigger chat interface, smaller text around it. "get eyes to the text box". if you have control on the color of the textbox, a small delta from the background color would make it obvious to me what to do next. note both cases of chat box more primary and color differentiation here.

[derrekcoleman]

@andrewxhill wdyt? I'm reluctant to double the height of the box when we don't have model toggle, photo upload, or other buttons asking for it.

[andrewxhill]

Lgtm

[vercel]

Additional Suggestion:

The client-side stream parser will crash with a JSON parsing error when it encounters the data: [DONE] marker that the OpenAI API sends to signal the end of the stream, causing the entire response to fail with "Sorry, something went wrong."

View Details

📝 Patch Details

diff --git a/components/ai/engines/openai.ts b/components/ai/engines/openai.ts
index 8e47d93..92f7b5d 100644
--- a/components/ai/engines/openai.ts
+++ b/components/ai/engines/openai.ts
@@ -89,7 +89,15 @@ export async function createOpenAIEngine(): Promise<Engine> {
         for (const line of lines) {
           if (aborted || !line.trim()) continue;
 
-          const json = JSON.parse(line);
+          // Handle Server-Sent Events format
+          let jsonData = line;
+          if (line.startsWith("data: ")) {
+            const data = line.slice(6);
+            if (data === "[DONE]") continue;
+            jsonData = data;
+          }
+
+          const json = JSON.parse(jsonData);
           if ("choices" in json && Array.isArray(json.choices)) {
             const delta = (json as OpenAIResponse).choices[0]?.delta?.content || "";
             if (delta) {

Analysis

Client-side SSE parser crashes on OpenAI's data: [DONE] marker

What fails: The JSON.parse(line) call in components/ai/engines/openai.ts:92 attempts to parse Server-Sent Events formatted lines directly as JSON, causing crashes on both valid SSE messages and the data: [DONE] termination marker.

How to reproduce:

 ]}" 
# "data: [DONE]"
# Both cause JSON.parse to fail since they include the "data: " SSE prefix

Result: JSON.parse("data: [DONE]") throws SyntaxError: Unexpected token 'd', caught by the try-catch block at line 114, replacing successful AI responses with "Sorry, something went wrong."

Expected: Client should handle SSE format correctly by parsing only the content after "data: " prefix, as the server-side transform already does at app/api/chat/route.ts:179-182