AI & Access Control全新
AI in a CRM usually comes as a black box: nobody knows what was sent to the model, how much each person spent or who may act on the company's behalf. SellioCRM puts those controls in the hands of administrators, under Settings. Under Artificial Intelligence you connect your own provider and key if you want, and turn on an audit trail of every call, personal data redaction before text leaves your account, and agent autonomy with a daily action cap. Under Corporate AI you define the language, tone and guidelines every AI call in the product carries, and keep a library of approved prompts. Under AI usage you see how much each person consumed this month and set credits per user or role. Under Global governance you centralize required fields, approval and retention per organization or business unit. And under Support access you decide when, and for how long, our support may see your data, with an immutable trail of every read and write.
- Bring your own AI key (provider and model of your choice) or use the system AI, metered against your spend cap
- AI governance: audit trail of every call, personal data redaction before the provider, and agent autonomy with a maximum of actions per day
- Corporate AI: response language, tone and guidelines applied to every AI call in the product, plus a library of approved prompts
- AI usage by user: monthly spend per person in dollars and a monthly credit per user or per role, refused once reached
- Consented support access: 24 hours, 72 hours or 7 days, revocable instantly, with a read, write and emergency trail nobody can erase
在 Sellio 中免费，你只需为机器付费
All of these controls are part of the free SellioCRM, as long as your account's infrastructure usage always stays under US$ 20 a month. System AI is metered against your usage and respects your spend cap; with your own key, AI usage runs on your provider account.
Personal data does not travel for nothing
With redaction on, emails, phone numbers, tax IDs and card numbers typed by people become markers before the call leaves. It is not a request to the model: the data never reaches it.
AI budget under control
You see who spent how much this month and set a credit per person or role before an experiment eats the team's budget.
AI with the company's voice
Language, tone and guidelines apply to every feature at once. A draft requested in Portuguese comes out in English, in the agreed tone, with no promised deadline.
An agent on a leash
Without autonomy, every write action asks for confirmation. With autonomy, there is a daily cap and a ledger of every action, so you know what the agent did and why.
Verifiable trust in support
Access is temporary, consented and revocable, and the trail separates reads from writes. If something unexpected shows up, you see it and ask.
Personal data redaction that is data that never leaves, not an instruction to the model: emails, phone numbers, tax IDs, SSNs and card numbers become markers before the call
Agent autonomy off by default: write actions ask for confirmation; on, they run up to the daily cap and every action goes to an append-only ledger
The AI audit trail stores prompt, response, model and tokens in an append-only ledger, and never stores cost
Support has no access by default: business records stay out of reach until an administrator turns access on, and it is the system that refuses, not a promise of good conduct
Your key, your safety rules
Under Settings, Artificial Intelligence, choose the provider and model and paste your API key: it is used only on the server and never shown again. With your own key, AI runs on your provider account. In the same place, the AI governance card turns on three brakes: keep an audit trail of every AI call, redact personal data before it reaches the provider, and allow the agent to act without confirmation, with the maximum actions per day you set. Only administrators can change these, and every change is recorded in the organization's audit trail.
One behavior policy for all AI
Under Corporate AI, turn on the policy and set the response language, the tone and free-text guidelines, such as never promising a deadline or a discount. From then on, every AI call made by the product carries those instructions ahead of the request: assistant, suggestions, agents and every other feature that generates text. The Allow customer data and Allow financial data switches are instructions to the model, not technical blocks; to keep data from leaving, use personal data redaction and field-level permissions. The prompt library keeps the requests that work for the team to copy.
How much each person spends, and how far they can go
Under AI usage, pick the month and see each person's consumption in credits (US$), with a total line. In the Monthly credit limits card, set a credit per user or per role; both coexist, and reaching either one causes that person's next calls to be refused with the limit notice until the month rolls over. Anyone without a limit is not limited. Only usage on the system AI key counts; with your own key, consumption belongs to your provider account.
Support only comes in when you open the door, and everything is logged
Without an active access, SellioCRM support sees only operational account information; contacts, companies, opportunities, proposals and conversations stay out of reach. Under Support access, an administrator picks the duration (24 hours, 72 hours or 7 days), notes the ticket reference and turns it on; access expires on its own and the Revoke access button cuts it instantly. The Access trail shows when, which agent, which action (read, write or emergency with a mandatory reason) and on what, and cannot be edited or deleted by anyone.
- 1
Under Settings, Artificial Intelligence, connect the provider and key, your own if you want, and in the AI governance card turn on the audit trail, personal data redaction and agent autonomy with its daily cap.
- 2
Under Corporate AI, enable the policy, set language, tone and guidelines, and register approved prompts in the library.
- 3
Under AI usage, follow this month's spend per person and set monthly credits per user or per role.
- 4
Under Global governance, create policies for required fields, approval and retention with Global or Business unit scope and check the effective policy.
- 5
Under Support access, turn access on only when there is a ticket, with duration and note, follow the Access trail and revoke when done.
Regulated company
Personal data redaction on, AI audit trail active and a corporate policy forbidding customer data in answers: every call is recorded with prompt, response and model, with no raw personal data.
Large team with AI for everyone
A monthly credit of US$ 25 for the Sales rep role and US$ 120 for the content person; on the 19th a rep hits the limit and only that rep is refused until the month rolls over.
Support ticket
The administrator turns access on for 24 hours with the ticket note, support investigates in the afternoon, the trail shows three reads, and the next day the admin revokes before it expires.
An agent working alone, within limits
Autonomy on with a cap of 50 actions per day: the agent runs the proposed writes up to the cap, each one lands in the ledger; on action 51 it stops and says why.
Do I need my own AI key?
No. Without your own key, the system AI is used, metered against your usage and respecting your spend cap. With your own key, AI runs on your provider account and that usage does not appear under AI usage.
Is personal data redaction an instruction to the model?
No. It is a substitution in the text before the call: emails, phone numbers, tax IDs, SSNs and card numbers become markers. The data never reaches the provider.
What does the AI audit trail keep?
Prompt, response, model and tokens of every call, in an append-only ledger. Cost is never stored. It starts off; an administrator turns it on under Artificial Intelligence, AI governance.
Is the monthly AI credit an absolute barrier?
It is a budget brake for normal use. The check runs before AI calls made by the product and, once exceeded, refuses with the limit notice; it was built never to bring AI down because of a failure in the check itself.
Can support come in without me turning access on?
Only in an emergency, with a mandatory written reason, and the event appears in the Access trail marked as emergency. There is no silent path, and the trail cannot be edited or deleted.