Brilo AI distinguishes between a Prompt and Knowledge to separate what the voice agent is told to do in the moment from the factual content it can draw on. A Prompt is the instruction set or template that shapes a single response or dialog turn. Knowledge is the indexed content Brilo AI searches to ground answers across calls, such as FAQs, policy text, or CRM fields. Together they control answer quality, retrieval behavior, and how the Brilo AI voice agent stays on-brand and accurate during conversations.

What’s the difference between Prompt vs Knowledge?

A prompt tells the agent how to respond; knowledge supplies the facts the agent can use.

How do prompts and knowledge work together in Brilo AI?

Prompts instruct the Brilo AI voice agent how to query and present content from your knowledge.

Is knowledge the same as training data in Brilo AI?

Not exactly; Knowledge is used for retrieval and grounding, while training data is used for model updates or supervised fine-tuning when applicable.

Enterprise buyers ask this because misconfiguring prompts or mixing ephemeral instructions with persistent facts creates inconsistent caller experiences. Financial services, banking, and healthcare teams need predictable voice agent outputs for regulation and auditability. Understanding the split between prompts and knowledge helps teams set guardrails, control escalation, and meet internal quality standards without exposing sensitive sources or over-triggering human handoffs.

When a caller asks a question, Brilo AI first evaluates the active Prompt to decide tone, required data fields, and allowed actions. The Prompt directs retrieval logic—what to search, how many knowledge snippets to return, and how to format the reply. Brilo AI then searches the configured Knowledge sources to ground the response before returning or executing the action.

In Brilo AI, Prompt is the runtime instruction set that defines response style, allowed actions, and retrieval parameters.

In Brilo AI, Knowledge is the indexed set of documents, FAQ entries, and structured fields the agent queries to produce factual answers.

See Brilo AI’s description of how agents learn from conversation context for more on runtime behavior: Brilo AI self-learning AI voice agents

Related technical terms: prompt template, retrieval, knowledge base, grounding, context window, answer quality.

Brilo AI enforces safety and quality through explicit guardrails in Prompts and by limiting Knowledge scopes. Prompts should include constraints such as “do not cite personal data” or “always offer human transfer for policy questions.” Knowledge sources can be restricted by namespace or document type to avoid exposing unapproved content.

Answer quality controls are the configured limits and fallback rules that decide when to use knowledge, when to ask clarifying questions, and when to escalate to a human. Prompts should set confidence thresholds and allowed actions; Knowledge should be reviewed and versioned so the agent never cites stale or unapproved documents.

For practical guidance on deflection and when to escalate, see Brilo AI’s call-deflection patterns: Brilo AI call deflection and routing examples

Healthcare example:

A hospital configures Knowledge with procedure descriptions and insurance coverage rules; the Prompt instructs the Brilo AI voice agent to never provide clinical advice and to escalate any symptomatic descriptions to a nurse. Knowledge provides the non-clinical facts the voice agent can read aloud, while the Prompt enforces clinical guardrails.

Banking / Financial services example:

A bank loads product terms and rates into Knowledge; the Prompt templates require the agent to confirm identity steps and to route any requests for account closures to a human. Knowledge supplies the factual terms and rates; the Prompt ensures compliance steps run before actions.

Insurance example:

An insurer uses Knowledge to store policy language and claim status. The Prompt instructs the Brilo AI voice agent to surface only policy summaries and to present a “request human review” option for contested claims.

(Do not treat these examples as legal or compliance advice. Map your Knowledge and Prompts to your own regulatory controls.)

Brilo AI voice agent workflows can escalate when prompts detect triggers such as low retrieval confidence, policy-sensitive content, or explicit caller requests. Typical handoff flows include:

Brilo AI handoffs are workflow-driven: configure the Prompt to call your CRM, open a ticket, or perform a warm transfer when escalation conditions are met.

If you want background on conversational design and template examples, review Brilo AI’s conversational AI overview: What is conversational AI and how Brilo AI designs conversations

Separating Prompts from Knowledge reduces maintenance overhead and improves governance. Teams can update facts in Knowledge without touching Prompts, and they can refine Prompts to change tone or escalation behavior without re-indexing content. This separation supports auditability, faster policy updates, and more consistent experience in regulated environments such as healthcare and banking.

Realistic outcomes include fewer agent escalations for routine factual requests, more controlled escalation when human review is required, and faster updates to facts or scripts under change control.

What is a Prompt in Brilo AI?

A Prompt in Brilo AI is the runtime instruction or template that defines how the voice agent should respond, which retrieval rules to use, and what escalation steps are allowed.

What is Knowledge in Brilo AI?

Knowledge is the set of indexed documents, structured fields, and approved content the Brilo AI voice agent searches to ground its answers and cite facts during calls.

Can I update Knowledge without changing Prompts?

Yes. Brilo AI is designed so you can update the knowledge base content independently of your Prompts, allowing you to refresh facts quickly while keeping dialog logic stable.

How do I prevent Brilo AI from exposing sensitive information?

Use Prompt constraints to block specific data types and scope Knowledge sources to only approved documents. Also configure confidence thresholds so the agent asks for human verification before disclosing sensitive items.

When should the agent escalate to a human?

Escalate when retrieval confidence is low, when a knowledge document contains a compliance flag, or when the Prompt detects a policy-sensitive utterance. Configure those triggers explicitly in Prompts.

Next actions: draft your initial Prompt templates, assemble a first Knowledge upload, and run a pilot call list to validate retrieval and handoff behavior.