From e87d19432b8bfd74aac68653a3da1c1485270db5 Mon Sep 17 00:00:00 2001
From: Peter Howell <peter.howell@gmail.com>
Date: Fri, 29 Aug 2025 22:35:35 +0000
Subject: [PATCH] gpt

---
 gpt.py | 59 +++++++++++++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 50 insertions(+), 9 deletions(-)

diff --git a/gpt.py b/gpt.py
index 884a066..9f6d90a 100644
--- a/gpt.py
+++ b/gpt.py
@@ -17,20 +17,61 @@ SAVE_ATTACHEMENTS = 1
 
 
 def gpt_chat(instruction, prompt, model=DEFAULT_MODEL):
+    # 1) Strip extremely long Outlook protection URLs first
+    try:
+        prompt = re.sub(r"\S*protection\.outlook\.com\S*", "", prompt, flags=re.I)
+    except Exception:
+        pass
+
+    # 2) Crude token estimation and truncation (target ~29k tokens)
+    #    Approximates 1 token ≈ 4 characters.
+    def _est_tokens(s):
+        try:
+            return max(1, int(len(s) / 4))
+        except Exception:
+            return 1
+
+    MAX_TOKENS = 29000
+    # reserve some tokens for system/overhead
+    overhead = 200
+    inst_tokens = _est_tokens(instruction)
+    prompt_tokens = _est_tokens(prompt)
+    total = inst_tokens + prompt_tokens + overhead
+    if total > MAX_TOKENS:
+        allow = max(500, MAX_TOKENS - inst_tokens - overhead)
+        # convert back to characters
+        allow_chars = max(2000, allow * 4)
+        # keep last N chars under assumption latest content is most relevant
+        if len(prompt) > allow_chars:
+            prompt = prompt[-allow_chars:]
+
     messages = [
         {"role": "system", "content": instruction},
         {"role": "user", "content": prompt}
-        ]
+    ]
 
-    response = client.chat.completions.create(model=model, messages=messages)
-    result = response.choices[0].message.content
+    try:
+        response = client.chat.completions.create(model=model, messages=messages)
+        result = response.choices[0].message.content
+    except Exception as e:
+        # Graceful failure; log and return empty JSON
+        try:
+            log3 = codecs.open('cache/gptlog.txt','a','utf-8')
+            log3.write(json.dumps({'prompt': prompt[:2000], 'error': str(e)}, indent=2))
+            log3.write("\n\n---\n\n")
+            log3.close()
+        except Exception:
+            pass
+        return "{}"
 
-    
-    log3 = codecs.open('cache/gptlog.txt','a','utf-8')
-    log3.write(json.dumps({'prompt':prompt, 'result': result},indent=2))
-    log3.write("\n\n---\n\n")
-    log3.close()
-    
+    # Log success
+    try:
+        log3 = codecs.open('cache/gptlog.txt','a','utf-8')
+        log3.write(json.dumps({'prompt':prompt, 'result': result},indent=2))
+        log3.write("\n\n---\n\n")
+        log3.close()
+    except Exception:
+        pass
     return result
 
 def summarize_u_info(msg):