EN
Idioma · Language · Lingua
AI security: a bounded, traceable mandate.
Defined once, restricted anywhere, never expanded.
An Abbi Action is a named operation that, given the same inputs, produces the same result and leaves a record. Abbiamo.ai defines it with its maximum limits; each following level can only keep it the same or make it narrower.
-
«Actions» console · Abbiamo.ai
What exists and how far it goes
Inputs and outputs, connector with encrypted credentials, validations, behavior on error, maximum limits, maximum autonomy, version and audit.
-
Console 1 · Entity
The autonomy ceiling
Set by the service owner. In Caddie it is fixed: he prepares, he does not execute.
-
Console 2 · Training
What is enabled
Which «Actions», when, and with equal or lower autonomy and limits.
-
Console 3 · Deployment
In what context
The deployment's own inputs, channel, hours and limits.
-
Console 4 · Evaluation
Nothing changes on its own
It reads the log and proposes adjustments; a person approves them.
-
Effective permission
The minimum of all
If the «Action» allows 50,000, the profile 20,000 and the deployment 10,000, the commitment is 10,000. One parameter, one owner.
Meet Abbi Studio →
What is said is what gets saved.
-
Mode
Prepare
Assembles the validated operation without writing to business systems. It is Caddie's mode, and Abbie's when she only suggests.
-
Mode
Execute
Writes to the system and returns a verifiable output. An «Action» that only executes cannot be enabled in Caddie.
-
Mode
Read-only
Queries and calculations, such as a balance or an installment, do not write in any mode.
-
Contract
Contract sheet
Every «Action» has its acceptance test: what is said matches what is saved, and prepare mode does not write.
-
Versions
Pinned version
Each profile is pinned to a version. If the «Action» contract changes, the profile that adopts it is revalidated.
-
Autonomy
Action by action
Suggests, executes with confirmation or executes on its own: it is decided for each «Action», not for the whole agent.
Explore the actions →
What no process can relax.
-
Lock
Hiding that it is AI
Abbie identifies herself as artificial intelligence and never poses as a human.
-
Lock
Stating without a source
Nothing that is not in an authorized, current source.
-
Lock
Making up conditions
No prices, balances, rates, deadlines or offers, and no new policies.
-
Lock
Asking for credentials
Never passwords, PINs, CVVs or tokens.
-
Lock
Applying pressure
No contacts outside permitted hours, no pressure, no threats.
-
Lock
Learning on her own
She improves only when a person approves a learning.
These rules are set in the entity, and no client, deployment or configuration level can relax them.
At any level, the same rules for your data.
-
Commitment
Minimization
Only the data the process needs is requested and used.
-
Commitment
Encryption
In transit and at rest; connector credentials are encrypted and never displayed.
-
Commitment
Retention and deletion
Configurable by the client; 180 days by default.
-
Commitment
One environment per client
In production, each client has its own «tenant».
-
Commitment
Full isolation
The universal knowledge base, each client's knowledge base and its transactional databases are isolated from one another. No client's data is visible to another.
-
Commitment
No training for others
A client's data does not train models for anyone else or improve someone else's product without their explicit authorization.
Where your data lives is decided with the sovereignty level: abbiamo.ai/soluciones/soberania-de-datos.
Every change has a name, a date and an approval.
-
Access
One user per person
Each person signs in to Abbi Studio with their own user; accounts are not shared, because every change is recorded under the name of whoever made it.
-
Cascade
No one relaxes what is above
Configuration flows down from the entity to the deployment, and no lower level can relax what the level above sets.
-
Learning
Improves only with approval
Every learning moves from proposed to approved or discarded by a person with permission on the corresponding console.
-
Versions
Automatic history
Date, author and diff of every change, with controlled A/B tests.
-
Alerts
From the first violation
A single violation of a rule is enough to trigger an alert, as is a drift in tone, a complaint or an «Action» error.
-
Human review
A sample scored by people
At least 2% of conversations, with a minimum of 50 per week, prioritizing escalations and alerts.
Abbi Studio design requirements. The scope of each implementation is confirmed in its contract.
What Abbiamo.ai protects and what your company controls.
Responsibility is shared and explicit: each party knows its part from the design stage.
- Abbiamo.ai protects
- The entity, the Abbi Studio consoles, the action layer, the encryption of connector credentials, the separation per client and the decision log. It retains the transcripts and the action logs.
- Your company controls
- Audio recording, the telephony and infrastructure it provides, the authorized sources, the mandate and each person's permissions.
- Defined jointly
- The sovereignty level, retention, who audits and how the right to erasure is exercised.
The exact division is set in the contract of each implementation.
Security, standards and data.
-
AI strategy
The why comes first
The belief, the six hows with their proof, what we sell and how to start with a first flow.
See the strategy → -
Regulation and compliance
Built for those who are held accountable
The standards we follow from the design stage, those of each industry and the decisions that always belong to your institution.
See the standards → -
Data sovereignty
Your data, where your case requires
Four deployment levels, from the cloud to your premises, and five questions to choose.
See the levels →
Let's review the security of your first flow.
Your request goes to the partner assigned to your company.
Let's talk about your operations →or write to us at hola@abbiamo.ai