Platform · AI agent
An agent that checks your system before answering.
Its own knowledge base, tools wired to your API and handoff to a human when confidence drops. Billed per interaction, no per-seat subscription.
Open in the panel
✦Billing Agent3 tools · knowledge base with 412 documentsactive
Hi, how much is my July invoice? Can I pay by Pix?
Tool220 ms
Looking up the customer's invoice
GET /billing/invoices?doc=***4410
→ 200 {total: 284.90,
due: "2026-08-05"}
Knowledge base180 ms
Pix payment rule · política-cobranca.pdf, p. 4
Agent reply1,4 s
Your July invoice is R$ 284,90, due on 05/08. You can pay by Pix — I just generated the code, valid for 30 minutes.
1 interaction · R$ 0,35Tools mode. The channel message is billed separately.
46%
of conversations resolved with no human touch
1,4 s
average reply with a call to your API
R$ 0,08
per simple interaction
4
billing modes, no seats
How the agent operates
Bounded by design, not by promise.
It answers from your knowledge base citing the source, calls your API when it needs real data, and hands off when the topic leaves the authorized scope. No improvising with your customer's money.
▤
Your knowledge base, your truth
Documents and URLs indexed. The reply cites the source and admits when it does not know.
⌘
Tools on your API
The agent looks up orders, invoices and customer records in real time, with minimal scope.
⇄
Handoff with a summary
Low confidence or a blocked topic: it transfers to the human queue with the case summarized.
◱
Cost per interaction
No per-seat subscription. You pay for the cycle the AI actually ran.
01Knowledge base
It answers from your documentation, not from the internet.
Upload PDFs, spreadsheets and help articles, or point to a URL. The agent fetches the relevant passage, answers citing the source and — when it finds nothing — says it does not know and hands off.
01Formats your operation already hasPDF, DOCX, CSV, text and web pages indexed automatically.
02Reply with a sourceThe passage used shows in the log, so you can audit where the information came from.
03Silence beats inventionWith no relevant passage the agent takes no chances: it routes to a human.
Knowledge base4 sources
▤politica-cobranca.pdf18 pages · updated 2 days agoIndexed
▤faq-trocas-e-devolucoes.docx42 questionsIndexed
◎ajuda.suaempresa.comURL · weekly recrawlSyncing
⊞tabela-fretes-2026.csv1.208 rowsIndexed
02Function calling
The agent queries your system in real time.
Each tool is one of your endpoints, with a declared parameter schema. The agent decides when to call it; you control what it can see and do — read only, or write as well.
01Minimal scope per toolIts own authentication and explicit read or write permission.
02Timeout and fallbackIf your API is slow or fails, the conversation goes to the human queue.
03Auditable callParameters sent and the response received are recorded in the interaction.
Declared tools4 endpoints
GETconsultar_fatura
https://api.suaempresa.com/billing/invoices
Read only. Takes the customer's tax ID and returns the amount and due date.
GETrastrear_pedido
https://api.suaempresa.com/orders/{id}/tracking
Returns status and delivery estimate for the agent to relay to the customer.
POSTgerar_pix
https://api.suaempresa.com/payments/pix
Authorized write, with an amount ceiling and an audit record.
POSTabrir_ticket
https://api.suaempresa.com/support/tickets
Used on handoff: creates the ticket with the case summary.
03Control
The AI operates within what you authorized.
Blocked topics, discount ceilings, mandatory human confirmation on sensitive actions and a tone of voice defined in writing. Anything not authorized becomes a handoff, not an improvisation.
01Blocked topicsLegal matters, regulators and negotiation outside policy fall out of scope.
02Autonomy ceilingDiscounts and deadlines up to the defined limit; above that, human approval.
03Privacy by designPersonal data masked in the log and nothing used to train a model.
Agent limitsalways enforced
✕Blocked topicsLegal matters, complaints to a regulator and negotiation outside policy become an immediate handoff.
⛊Autonomy ceilingDiscounts and extensions only up to the defined ceiling; above that, human confirmation is mandatory.
✎Tone of voice in writingStyle, form of address and banned words applied to every reply.
◐Privacy by designPersonal data masked in the log and no conversation used to train a model.
04Traceability
Every reply has a record.
The log keeps the question, the reply, the passages cited, the tool calls, the mode billed and the cost. You can filter by period, queue, agent and outcome — and export.
01Cost line by lineThe billed mode shows on every interaction, with no reconciliation afterwards.
02Review that improves the baseFlagging a reply as incorrect puts it in the content review queue.
03Export by periodFilter by queue, agent and outcome, with a download for auditing.
Interaction logexportable
14:03ToolHow much is my July invoice?R$ 0,35
14:01RAGHow does exchanging a defective product work?R$ 0,20
13:58SimpleAre you open on Saturdays?R$ 0,08
13:56HandoffI want to file a formal complaintR$ 0,08
13:54AdvancedCompare the three proposals I attachedR$ 0,60
Cost per interaction
You pay for the cycle the AI ran.
The mode is set by the most advanced capability the interaction used. Simulating helps you understand the cost before you activate.
✓No per-seat subscription and no minimum commitment
✓AI billing adds to the channel's message price
✓Interactions count toward the tier volume measurement
ModePriceExample at 20,000
SimpleOnly the model replied, with no knowledge base and no toolsR$ 0,08R$ 1.600,00
Knowledge base (RAG)Queried your knowledge base and used the result in the replyR$ 0,20R$ 4.000,00
Tools (function calling)Ran an operational action: look up an order, a profile, an integrationR$ 0,35R$ 7.000,00
AdvancedAdvanced model enabled or tokens above the thresholdR$ 0,60R$ 12.000,00
Multimodal (vision)Interaction that processes an image — priced according to the model usedon requestbased on usage
Interaction simulator
Interactions per month40k/month
Simple R$ 0,0845% · R$ 1.440,00
Knowledge base (RAG) R$ 0,2035% · R$ 2.800,00
Tools (function calling) R$ 0,3515% · R$ 2.100,00
Advanced R$ 0,605% · R$ 1.200,00
Estimated total per month
R$ 7.540,00
R$ 0,189 per interaction
Simple18,000R$ 1.440,00
Knowledge base (RAG)14,000R$ 2.800,00
Tools (function calling)6,000R$ 2.100,00
Advanced2,000R$ 1.200,00
Where the agent already works
Cases the operation closes without calling anyone.
▣
Order status
Looks up tracking on your API and relays the delivery estimate.
68% with no human
◱
Invoice copy and Pix
Generates the code on the spot, with an amount ceiling and an audit trail.
R$ 0,35 / interaction
⇄
Debt negotiation
Offers installments within policy and hands off above the ceiling.
handoff on 12%
↻
Exchange and return
Answers from the policy, opens the ticket and schedules the pickup.
42% with no human
◎
Lead qualification
Asks the sales team's questions and books the meeting.
3× faster
◫
Support triage
Classifies, collects the screenshot and log and routes to the right queue.
SLA −38%
✦
Copilot for agents
Suggests a draft in the inbox; the human reviews before sending.
71% accepted
▣
Order status
Looks up tracking on your API and relays the delivery estimate.
68% with no human
◱
Invoice copy and Pix
Generates the code on the spot, with an amount ceiling and an audit trail.
R$ 0,35 / interaction
⇄
Debt negotiation
Offers installments within policy and hands off above the ceiling.
handoff on 12%
↻
Exchange and return
Answers from the policy, opens the ticket and schedules the pickup.
42% with no human
◎
Lead qualification
Asks the sales team's questions and books the meeting.
3× faster
◫
Support triage
Classifies, collects the screenshot and log and routes to the right queue.
SLA −38%
✦
Copilot for agents
Suggests a draft in the inbox; the human reviews before sending.
71% accepted
Activation
From toggle to first conversation in a day.
This is not an integration project. Upload the base, declare the tools and pick the mode.
10 minTurn on the serviceOne toggle on the organization enables the AI. No extra contract, no per-user license.
1 hourUpload the baseDrag in the documents or point to your help center URL. Indexing is automatic.
half a dayWire up the toolsDeclare the endpoints the agent can call, each with its own authentication and scope.
whenever you wantPick the modeStart as a copilot, suggesting to the human agent, and release automatic replies queue by queue.
Multiple agents
One agent per operation, not one generic agent for everything.
Each agent has its own knowledge base, tools and tone. The inbox queue decides which one handles each conversation.
✦Billing Agent3 tools · Billing queueactive
✦Sales Agentcatalog knowledge base · Sales queueactive
✦Support L1 Agentcopilot · suggests to the human agentcopilot
“AI took over 46% of our billing conversations and the human team kept what has margin: real negotiation.”
DPDiego PradoOperations Director · debt collection company
Questions about the AI
What people ask before switching the agent on.
Want to design it around your case? Talk to the team .
The risk exists in any system, and that is why the agent is bounded by design: it answers from your knowledge base citing the source, hands off when it finds no answer and requires human confirmation on sensitive actions. The recommendation is to start in copilot mode, where it only suggests and the human agent decides.
Start today with R$ 50 in test credit.
Sandbox wallet released when you create the account, a WhatsApp test number and guided onboarding. No card, no contract.
Brazilian CPaaS platform. WhatsApp, email, SMS, RCS, push and Instagram in a single integration.
All systems operationalDevelopers
Company
© 2026 CCX Digital Solutions LTDA · CNPJ 24.329.191/0001-06 · BrazilTerms of usePrivacySecurityChangelog