My staff are not technical. Will they really be able to use it?
Yes, because there is no new tool to learn. The assistant lives in the chats the team already uses — Slack, Discord, Telegram — and you write to it, or send it a voice note, the way you would to a colleague: "draft the email to Rossi about Friday's deadline". No interface to study, no technical skill required.
What does Cerase give me that an AI subscription for the team does not already?
An AI subscription provides a powerful tool, but one that each person has to open, learn and steer by hand, and that knows nothing about your company and does not talk to your Italian systems. Cerase is a different thing: assistants ready for each role, living in the chats where the team already works — no new app to adopt. They know your procedures and operate on your systems (ERP, certified email, invoicing) within the permissions you set. We configure them around the way you work; they stay yours and exportable, and they are not tied to a single AI supplier. In short: you are not adding one more chat licence, you are bringing AI into everyday work — compliant and under control.
Which business areas does Cerase work for?
Every area of an Italian SME. We provide ready-made presets for the most common functions — administration, sales, marketing, management — and calibrate them on your procedures, your data sources and your terminology. For any other area we build the assistant on the same foundation.
How do data, knowledge and configurations stay in my hands?
Three guarantees, written into the contract. One: the environment is dedicated to your company, on reserved European servers — your data is never mixed with anyone else's. Two: knowledge base, history, audit log, connectors and policies are exportable in standard formats at any moment, without having to ask. Three: Cerase is independent of the AI model working behind the scenes, so as models evolve your setup stays intact. Your work is, and remains, yours.
Is my data used to train the AI?
Identifiers relating to you, your clients and your company never leave our systems in the clear: before reaching the model they are replaced by placeholders, and they are reassembled only on our machines in the EU, which hold the only key for doing so. The model works on text with no identifiers in it, which it cannot trace back to you. The filter is configurable, so we extend the masking to the data that is critical in your sector. Everything that is yours — knowledge, history, configurations — stays in an EU environment, exportable and deletable whenever you want.
Can I decide what the assistant does on its own and what a person has to approve?
You decide, connector by connector, choosing the level: read only, draft only (prepares but does not send), with confirmation, or autonomous. For example: a salesperson can let follow-up emails go out on their own, while changes to contracts stay "with confirmation"; in administration, every action on an invoice goes through an approval. These are your company's settings, not defaults chosen by us, and you change them from the console whenever you like. Every conversation and every action is recorded in a complete audit log.
Does the assistant reply to incoming emails on its own?
No. It checks the mail when you ask it in chat, or at the time you have set, and summarises what has arrived. Sending follows the same rule as the other tools: above the threshold you decided, it asks for your OK before proceeding. Every message that goes out states that it was written by an AI assistant and carries your company's details, as the law requires.
What does sending an email cost?
One credit per message sent by the assistant. Receiving costs nothing, and neither do the platform's service emails such as invitations. The credit is there to put a ceiling in place: an assistant must not be able to send hundreds of messages without you noticing.
Can the assistant see or reveal the connector credentials?
No. The assistant and the secrets live in separate containers: tokens, passwords and API keys are not in the environment the assistant runs in and never enter the prompt. They are held by the connector Gateway — an isolated service — which carries out only the actions the admin has authorised. The assistant can ask to "send this email" or "update this record", but it cannot read or display the credentials that make the action possible.
How do you guarantee a skill contains no dangerous instructions?
Every skill is analysed automatically before it can be enabled: the check flags risky instructions, attempts to get around permissions, or behaviour outside the agreed perimeter. The admin sees the result and decides; no procedure comes into service without having been enabled on the template. Even once active, a skill stays bound to the connector's permissions and to the human approval levels you have set.
How much work is it to integrate Cerase with the ERP and CRM we already use?
Not much. Cerase connects to your systems the way a new user would, with the permissions you assign it: no data migration, no workflow to rewrite. The integrations most common in Italian SMEs — FatturaPA, certified email, the widely used ERP systems, Google Workspace, Microsoft 365, the main CRMs — are ready to use; the more specific ones are configured by one of our integration partners during onboarding.
From signing, when do we see the first result in the company?
Soon. The technical setup — connectors, permissions, first template per role — takes 2-3 days, and the first person on the team sends their first real action by the third day. Tuning to your specific processes continues in parallel over the following 1-2 weeks, while the team is already using the assistant on standard cases. No data migration; any work on dedicated connectors stays targeted and contained.
Who handles the setup?
There are two routes, and you choose the right one. Many configurations you handle on your own, following our guide; when you would rather be accompanied, we point you to the integration partner best suited to your case, on quotation.
What support is included?
Every mode includes
Standard support: a support contact who knows your configuration — not an anonymous channel — with continuous operational support, monitoring of the environment and the response times set out in our SLA, the same across all modes. For more structured needs there is
Premium support (on quotation): a dedicated contact, a priority channel, periodic optimisation sessions, continuous training for the team and value reporting.
See service and guarantees.
What service levels do you guarantee?
The SLA is the same in every mode: service availability of
99.5% on a monthly basis and response within
4 working hours if the service is unavailable (within 1 working day for degradation of a significant function, within 2 for non-blocking requests). On top of that: daily backup with restore within 1 working day, encrypted copies kept in the EU for up to 24 months, proactive notification of incidents, and the state of your environment always available to check in real time.
See all the details.
How is plan consumption calculated?
Credits are the unit Cerase measures consumption in. On Credits included plans you have a monthly allowance already covered by the fee; if the monthly pool runs out you can buy a pack. In Pay as you go you have a minimum threshold included and buy packs when you need them. In every mode, credit packs also cover paid Cerase add-ons such as connectors, extra skills, searches and attachments. On plans using models supplied by Cerase, the assistant's answers, attachments and web searches all draw on the same pool. What a task consumes depends on how much text the assistant has to read and produce to finish it, and on the plan's model: Base uses the lighter model and consumes less for the same work. Web searches and attachments — OCR, audio transcription, reading images — consume credits too. A task repeated on the same material consumes far less than the first time, because the context already processed is reused from cache.
What is the Pro plan's boost?
It is a switch in the console: you turn it on for a single assistant and, for as long as you decide, that assistant moves to the most powerful model, then returns to the plan's model on its own. It consumes six times the credits the same work would consume without the boost, and takes them from the plan's allowance: it is not a separate line on the invoice. The other assistants do not change model and do not consume more.
What happens if I run out of credits for the month?
You keep control throughout, with no surprises on the invoice. Credits an assistant does not use flow into the company's shared pool, and you decide — assistant by assistant — who may draw on it once past their own monthly threshold. Anyone not enabled stops at the threshold and tells you, with no unexpected cost. If the pool runs out as well, or you need credits for Cerase add-ons such as connectors and extra skills, you can buy packs — 5,000 (€29), 50,000 (€249), 100,000 (€449), 200,000 (€799), valid for 12 months. An assistant with the boost switched on that draws on the pool beyond its own threshold consumes at a higher rate.
Can I use my own AI provider (BYOK)?
Yes. Cerase can operate with your own model provider: you can connect the key for your cloud account — Mistral (European), Azure OpenAI in the EU, or Anthropic, for instance — or point Cerase at a model running on your own infrastructure, including locally (with Ollama, for example), as long as it exposes an OpenAI-compatible API. Either way the model, the tokens and the data stay inside your perimeter, at your rate, while Cerase provides the platform, the integration and the governance. This is the right configuration when you already have an AI supply arrangement, negotiated rates, or requirements that keep model and data in-house.
See the BYOK mode in pricing.
