Your key, 8 providers
OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI, DigitalOcean Gradient or any compatible URL. The key belongs to your company, and it leaves whenever your company wants.
AI agent The AI is yours.
The AI is yours.
bFocus just puts it to work
Your key, your model, your depth and your cap in reais. We do not charge per resolution and we put no markup on tokens. With no key registered, bFocus works exactly the same — just without AI.
- 8 providers
- Cap in R$ at the PTAX rate
- Switching it off is a legitimate choice
Who is in charge Four decisions that
Four decisions that
are yours alone
Your model
bFocus reads the models your key can reach, groups them by tier and recommends one, showing the price of the model you pick. Switching to something cheaper or stronger is one click.
Your depth, per product
None, assisted or autonomous. Assisted suggests to the agent; autonomous talks to the customer and hands over when it has no source.
Your cap in reais
Per conversation, per user per day and per product per month. Conversion uses the Central Bank of Brazil PTAX rate and the call is blocked before the cap is breached.
How it answers With a source,
With a source,
or it does not answer
The rule that heads off the biggest risk of AI in support: the invented answer.
- 1
It searches your base
Hybrid search across the articles for that product — by text and, when your key generates vectors, by meaning too — plus the description you wrote for the AI.
- 2
It answers only from an article
The answer has to cite an article from the base, or it is discarded. In assisted mode, the agent sees the source of each suggestion.
- 3
In doubt, it asks
If something is missing, it asks back instead of guessing — and tries again with what it got.
- 4
No source, it hands over
It goes to a human with a summary, the data collected and the reason for the handover written out in a message.
A key, network or cap failure also becomes a handover: the customer is never left without an answer.
Teaching the AI It gets better with what
It gets better with what
you have already written
The knowledge base is the fuel. bFocus helps you build it from the material your company already has — and always with a human review before publishing.
- Import .md files and turn them into articles by product
- A prompt kit with "copy and open in" ChatGPT, Claude or Gemini, to draft elsewhere and bring back finished
- The AI reads sites, URLs and files and shows the estimate in reais before it runs
- Everything arrives as a draft: nothing is published without someone reviewing it — and the agent only reads what is published, so remember to publish after importing
- An article generated from a real ticket, with sensitive data masked
- Publishing requires ticking "I reviewed this" — the responsibility stays with a person
Evaluation and gaps
A set of standard questions scores the agent: did it escalate when it should, did it cite a valid source, did it cover the expected points. And every unanswered question becomes a gap, which reopens if the published article did not settle it.
What it costs The bill with
The bill with
no toll booth
A thousand conversations a month, on a fast model, next to what the market charges per resolution.
- ~R$ 200
- 1,000 conversations at your own provider, on a fast model (estimate)
- R$ 0
- how much of those conversations goes to bFocus
- R$ 5.35
- per resolution on Intercom (public price)
- ~R$ 5.25
- per resolution on Octadesk (15 credits at R$ 0.35)
Estimate based on roughly 30,000 input tokens per conversation on a Haiku-class model; the real figure depends on the model you choose and the size of your base. Market prices checked in September 2026. Search by meaning is part of that same bill: the vectors for your base also come from your key. There is no exception — bFocus does not supply AI.
Safeguards What it
What it
never does
It does not invent
With no source in the base there is no answer: it either asks or hands over. On handover, the internal note lists the sources it checked.
It does not trap the customer
"Talk to an agent" is always available, as the Brazilian SAC rules require — and an agent can take over the conversation at any moment.
It does not ignore abuse
An attempt to manipulate the prompt or to steer into forbidden subjects is logged, along with what was blocked.
Questions about the AI
Do you train models on my data?
No. bFocus does not train any model. The conversation goes to the provider you chose, under the contract you hold with them, using your key.
What if I do not want AI?
Just do not register a key, or leave the level on "none". Everything else in the product works the same: queue, SLA, chat with people, knowledge base and indicators.
What happens when the cap in reais is reached?
The call is blocked before it spends anything, support carries on with a human, and the administrator sees what hit the limit — per conversation, per user or per product.
Can I use different models per product?
You can. The autonomy level and the agent are set per product, so you can leave a mature product on autonomous and another one on assisted only.
Does search over the base use my key as well?
It does. There is no exception: the vectors behind search by meaning are generated with your company key — OpenAI, Google, Mistral or your own DigitalOcean Gradient key — and that usage goes on your company AI bill. If your only key comes from a provider without embeddings, such as Anthropic, the base carries on with text search. Conversation, summary, classification and reply suggestions have always used your key.
Bring your key and watch it work
During the 14-day trial you register the key, set a low cap and watch the AI answer with a source.