Main limitations (in general):

- **Knowledge cutoff & freshness:** My built-in knowledge is only current up to **2025-08**. I may be unaware of newer events, releases, prices, policies, or research unless you provide details.  
- **No guaranteed accuracy:** I can make mistakes or “hallucinate” details (names, citations, steps, numbers). I’m best used with verification for high-stakes tasks.  
- **No direct real‑world access by default:** I typically can’t reliably **browse the web, query live databases, check real-time availability, or access private systems** unless the environment explicitly provides those tools/data.  
- **Limited context window:** I don’t retain everything forever within a long conversation; very long threads or large documents may exceed what I can keep in working memory at once.  
- **Not a professional substitute:** I can help explain and draft, but I’m not a replacement for **medical, legal, financial, or safety** professionals, and I can miss important nuances.  
- **Can’t perform actions in the world:** I can’t actually place orders, send emails, run code on your machine, or change settings—unless you do it or tools are integrated.  
- **Sensitivity to input quality:** Ambiguous prompts, missing constraints, or inconsistent requirements can lead to poor outputs; I may need clarifying questions.  
- **Bias & policy constraints:** I may reflect biases present in training data, and I must refuse or limit help for certain harmful/illegal requests.  
- **Privacy limits:** I don’t have access to your private data unless you share it, and you should avoid providing sensitive info you wouldn’t want stored/handled by a third party system.

If you tell me what you want to use me for (e.g., coding, writing, planning, studying), I can list the most relevant limitations for that use case and how to work around them.
