Show readers the exact instruction the model runs under, its tools and its scope — and optionally let them add their own.
How this works
An answer is easier to trust when the rules behind it are visible. With prompt.show, a How this works disclosure in the panel shows the system instruction, the tools the model may call, and what it is searching. With prompt.allowAppend, readers can add an instruction of their own.
DocPilot
How this works
Instructions
You answer questions about Acme using only the excerpts you are given. Never invent APIs, methods, fields or routes. Mark every claim with a citation marker…
Tools (delivered as tool definitions)
search_docs · fetch_section · expand_section · list_pages · answerScope
All documentation pages are searched.
Your instruction
None. Anything you add is sent with your question, never as part of the instruction above.
Ask about Acme
All docsAI-generated. Check the linked pages.
Options
| Key | Default | What it does |
|---|---|---|
prompt.show | false | Adds the disclosure: instructions, tools, scope. |
prompt.allowAppend | false | Lets the reader add an instruction. It is sent with the question, never inside the system instruction, and kept for the session only. |
prompt.appendMaxChars | 500 | Length limit for the reader's instruction. |
prompt.extend | '' | Text you add after the shipped instruction, at build time. |
prompt.override | null | Replaces the shipped instruction entirely. |
On a site with its own tools, the disclosure lists them with the built-in ones, with the descriptions the model reads, because it publishes what is sent.
More: prompt in the reference.