Direct Answer (TL;DR)
Brilo AI Knowledge Sources are the documents, systems, and live inputs Brilo AI voice agents consult when answering callers. Brilo AI can be configured to use uploaded knowledge base articles and FAQs, CRM records, call transcripts and historical interactions, real-time system data via your webhook endpoint, and indexed attachments (PDFs, DOCX) that you provide. Brilo AI applies relevance ranking (vector search and confidence scoring) to choose answers and can attach metadata such as source name and timestamp to each response. You control which sources are enabled, how often they refresh, and the escalation rules when confidence is low.
What sources does Brilo AI use for knowledge? — Brilo AI uses documents, CRM data, transcripts, and live system data.
Which knowledge sources back a Brilo AI voice agent? — Brilo AI uses uploaded KBs, CRM records, transcription archives, and webhooks.
How does Brilo AI find answers from my systems? — Brilo AI indexes your content (embeddings and vector search) and ranks candidate answers by relevance and confidence.
Why This Question Comes Up (problem context)
Enterprises ask this because knowledge source selection affects accuracy, compliance, and integration effort. Regulated organizations in healthcare, banking, and insurance need clarity on what systems the Brilo AI voice agent will read from, how live data is used, and where audit trails live. Knowledge source choices also determine who must approve content, how frequently updates run, and whether the agent can surface personally identifiable or sensitive data.
How It Works (High-Level)
When you enable Brilo AI Knowledge Sources, Brilo AI performs three core steps: ingestion, indexing, and retrieval.
During ingestion, Brilo AI accepts files, structured records, and streaming inputs you authorize.
During indexing, Brilo AI converts text into searchable representations (embeddings) and stores metadata so results can be ranked by relevance.
During retrieval, Brilo AI runs a relevance search (vector search) and applies answer generation and confidence scoring before returning a response.
In Brilo AI, knowledge source is any approved system or document that Brilo AI may consult to generate answers (examples: uploaded PDFs, FAQ pages, or CRM records).
In Brilo AI, knowledge refresh is the scheduled or manual process that re-indexes updated documents so the voice agent uses current content.
In Brilo AI, confidence score is a numeric indicator Brilo AI assigns to each candidate answer to decide whether to respond automatically or route to a human.
Technical terms included: knowledge base, embeddings, vector search, webhook, CRM, transcription, intent detection, confidence scoring.
Guardrails & Boundaries
Brilo AI enforces configurable guardrails so the voice agent only uses approved sources and only exposes allowed fields. Typical guardrails include:
Source whitelists and blacklists so only designated folders, documents, and CRM fields are searchable.
Field-level masking to prevent exposing PII or regulated fields unless explicitly permitted.
Minimum confidence thresholds that trigger escalation to an agent instead of answering automatically.
Read-only access to source systems unless you enable action workflows separately.
In Brilo AI, read permission scope is the access level you grant Brilo AI that limits which records and fields it can index or surface. Use these guardrails to keep the agent within compliance and operational boundaries.
Applied Examples
Healthcare example:
A medical practice configures Brilo AI with scheduling FAQs, secure appointment-status data from the practice management system (read-only), and recent call transcripts. The Brilo AI voice agent answers appointment time and location questions from the indexed FAQs and authenticated scheduling records, but when a caller asks for clinical advice, the agent escalates to a human clinician or records a callback request.
Banking / Insurance example:
An insurance call center loads policy documents, underwriting FAQs, and policyholder records from the CRM. Brilo AI pulls policy language to explain coverage definitions, verifies identity via CRM fields, and routes high-risk or low-confidence inquiries (possible claims or fraud) to specialist agents with the relevant context attached.
Note: Do not interpret these examples as compliance advice; configure your access and approvals to meet your regulatory requirements.
Human Handoff & Escalation
Brilo AI voice agent workflows support multiple handoff patterns:
Intent or confidence trigger: When intent detection finds a “claim” intent or confidence falls below your threshold, the agent routes to a live representative.
Contextual warm transfer: Brilo AI passes conversation context, last three transcripts, and source citations to the receiving agent or CRM ticket.
Deferred escalation: The agent creates a task or ticket in your system and schedules a human callback when no agent is available.
You configure escalation rules and the amount of context included. For sensitive situations, choose minimal context and escalate only identifiers needed to continue the conversation.
Setup Requirements
Gather and upload canonical knowledge documents (PDFs, DOCX, or plain text) and an FAQ inventory to seed Brilo AI’s knowledge base.
Grant read access to your CRM fields you want surfaced and use field-level controls to limit exposure.
Provide a webhook endpoint for real-time data or status checks if live system data is required.
Define routing, confidence thresholds, and escalation workflows in the Brilo AI console.
Run verification calls in a staging environment and review the agent’s source citations and confidence scores.
Sign off on the approved source list and refresh cadence before going live.
Business Outcomes
Using Brilo AI Knowledge Sources typically yields more consistent customer answers, faster call resolution for scripted queries, and reduced load on live agents for repeatable tasks such as appointment scheduling or policy lookups. Because Brilo AI surfaces source citations and confidence scores, teams gain traceability for audit and training, which improves answer quality over time without increasing staffing proportionally.
FAQs
Which file types can Brilo AI ingest as knowledge sources?
Brilo AI accepts standard text documents and structured content you upload (for example, PDFs, DOCX, and plain text). Confirm accepted formats and size limits with your implementation team.
Can Brilo AI access live account balances or policy data?
Yes—when you provide a secure webhook or grant read access to your CRM, Brilo AI can retrieve live fields to include in answers. Configure field permissions and masking to control which data the agent may surface.
How often does Brilo AI refresh indexed knowledge?
You set the refresh cadence. Brilo AI supports scheduled re-indexing and on-demand refreshes so updates to policy documents or FAQs become available according to your change control process.
What happens if the agent is unsure about an answer?
When confidence is below your configured threshold, Brilo AI follows the escalation rule you set: ask the caller a clarifying question, create a ticket, or transfer to a human agent with context and source citations.
Can I restrict the agent from using certain documents?
Yes. Brilo AI supports source whitelists and blacklists as well as folder-level controls so you can exclude any documents from indexing.
Next Step
Review your content: prepare a canonical set of knowledge base documents and an FAQ inventory to upload to Brilo AI.
Connect systems: provide read access to the CRM fields you want the agent to use and a webhook endpoint if you need live data.
Engage Brilo AI: contact your Brilo AI account team to request a staging setup and run the first knowledge ingestion and test calls.