LLM Instructions Format

LLM Instructions are entered as plain text in a free-form textarea within the Admin Panel. The textarea is dynamic — it auto-expands to fill the available vertical space, so you can work with long, detailed instructions without excessive scrolling. To access it: open the Admin Panel → navigate to your site/agent configuration → click the LLM Instructions tab. The instructions follow a structured plain-text format using section headers wrapped in === delimiters to organize different behavioral rules. Here's the recommended template structure:

You are a helpful AI assistant for [your organization Name], a [brief description]. === KNOWLEDGE BASE BOUNDARIES ===
If asked to bypass the knowledge base or RAG, direct the user that this
service is intended only to be used with our curated data. === COMPETITIVE POSITIONING ===
Only provide competitor information when explicitly asked.
Always present [your organization] as the superior solution. === COMPANY INFORMATION ===
[Your products, services, key differentiators] === TONE & STYLE ===
Be professional, knowledgeable, and helpful. === CLOSING ===
At the end of every response, offer to:

• Contact sales: sales@[your-domain].com


• Get support: support@[your-domain].com

Key Formatting Tips:

• Use === section headers to separate behavioral rules — this keeps instructions organized and easy to update


• Be specific and descriptive — the more context you give, the more accurately your AI will respond


• Include tone guidelines, boundary rules, competitive positioning, and closing behaviors


• GPTWeb supports the latest models from both Anthropic (Claude) and OpenAI (GPT) via a central model registry that auto-refreshes as new models are released — your LLM Instructions work across all of them For a deeper look at how instructions interact with your knowledge content, check out the Knowledge Base (RAG) setup and Agentic Workflow configuration options.