Public agent instructionsExternal · website visitors
This agent answers visitors from approved public sources only. It cannot reach private datasets and must verify identity before any account-specific answer.
How the agent introduces itself and where it is used.
Internal note only. This is not sent to the model.
Where visitors meet this agent, and what it may collect. Public identifiers are not secrets; authorization happens server-side.
Show citations to visitors
Link every factual answer to an approved public source.
Collect contact details
Only with disclosure and explicit consent; never pre-checked.
Offer human handoff
Escalate unresolved or account-specific questions to Chatwoot.
Enable public API access
Short-lived session tokens only; no backend secrets in the browser.
The public scope, tone, and customer-facing limits. Published as an immutable version.
What happens when the agent cannot ground an answer.
Handoff includes the visitor transcript and citations. Account data requires verified identity.
Require public citations
Abstain when no approved source supports the answer.
Enforced outside the model. Instructions cannot grant access that policy denies.
Approved public sources only
Identity assertion for account-level data
Never invent pricing, availability, or order status
Collect contact details only with consent
Escalate unresolved questions to Chatwoot
No private workspace data in public sessions
Shown to website visitors before they type. Keep them inside public scope.
Draft changes are stored separately from the published version.
Public identifiers are not secrets. Domain allowlists and rate limits reduce abuse but do not replace authorization.
DeepSeek Flash
$0.15 / $0.60 per 1M tokens
Lower values keep support answers consistent.
State the job, the limits, and the handoff. Avoid stacking rules that contradict each other — models follow the closest instruction unreliably.
- Name the business and the visitor audience.
- Say what to do when public evidence is missing.
- Put identity requirements in the tool policy, not the prompt.
- Keep fallback language in the customer's voice.
Prompt isn't a security boundary
Tool permissions, tenant filters, and public/internal visibility are enforced outside the model. Changing this text cannot grant access.