AcqoraGX
CRM + Dialer

It's on by default, and you can turn it off

"Help improve the AcqoraGX model" is on for every workspace unless the owner turns it off. The owner can switch it off at any time in Integrations → AI. Turning it off stops new capture immediately, and your workspace's examples and knowledge are left out of every training dataset built after that. Datasets are re-checked at build time, so a later opt-out still counts. Models that were already trained can't "unlearn" data, but they're replaced as newer versions are trained.

What we collect

• AI examples: the request your team sent to an AI feature, and the AI's answer, along with the task (prep, coach, follow-up and so on), the provider and the model.
• Your team's feedback: thumbs up or down, and any corrected answers your team writes.
• Knowledge items ("canon"): short facts, definitions, objection responses and procedures that your team approves, or that the system confirms when several independent, approved examples agree (see below).

We don't collect your contact lists, recordings or messages for training. Provider API keys are encrypted and are never part of any dataset.

Masking

"Mask personal info" is on by default. Before anything is stored, email addresses, phone numbers and street numbers are replaced with consistent tokens such as [PHONE_3fa2c1]. Workspace ids are hashed in exports. Masking uses patterns, so names and free-form addresses may not be caught. Avoid putting sensitive personal details into AI requests.

Self-learning by agreement

When your team approves an AI answer or a call's CRM updates, your own AI provider may suggest up to three reusable knowledge items. Agreement counts as evidence, not proof. A low-risk item (a fact, definition, objection response, common mistake or example) is added to your workspace's knowledge automatically only when a published, versioned policy is fully met. The current pilot policy requires all of these: at least two independent sources (copies of one record, or several summaries of the same call or the same AI answer, count as one source), observations on at least two different days, at least 3 approved examples from at least two different teammates, no open conflict, and sources your workspace is still allowed to use. An item that agrees with knowledge you already approved is recorded as extra evidence for that item. Every decision is logged in a tamper-evident history, versioned, and can be rolled back.

Procedures, qualification or decision rules, anything about pricing, offers, compliance or legal matters, and anything that mentions sensitive personal details always need an owner, admin or manager to approve them. So does anything that conflicts with existing knowledge: both versions are kept until a person decides. An approval records who decided; it isn't a guarantee that the item is correct. If a teammate later rejects or corrects something that an automatic item relied on, that item is withdrawn and flagged.

Being approved doesn't make knowledge training data. Knowledge that was added only because teammates agreed is used for your workspace's AI answers but isn't used to train shared AcqoraGX models.

Licensing rule

Answers written by Anthropic, OpenAI or Google models are never used as training targets, because their terms don't allow it. They're used only for evaluation and retrieval. The same applies to knowledge extracted from those answers, even when another model did the extracting. Training targets come only from three sources: answers your team wrote or corrected, approved knowledge written by people or by open models, and outputs of providers the operator has confirmed allow training (currently: OpenAI-compatible endpoints, AcqoraGX model).

Shared improvements

Your workspace's examples train shared AcqoraGX models only while contribution is on. Knowledge items are shared with other workspaces only if the owner sets the scope to "shared", and even then AcqoraGX staff review each one first. When we show that several workspaces agree on an item, we only show it once at least three agree, and we never say which workspaces they are.

Retention

Captured examples are kept for up to two years (configurable by the operator) and then deleted. Only the most recent training datasets are kept; older dataset files are deleted. To train and evaluate models, a copy of a dataset is sent to our training infrastructure: processing nodes operated by AcqoraGX or its service providers. They receive only the dataset itself (no provider API keys, contact lists or recordings), run only our own built-in training and evaluation programs, and are instructed to delete the dataset and working files when the job finishes. They keep trained model files and evaluation results for up to seven days. A node reports what it deleted. That report is not proof that every copy is gone, because backups and disk blocks may persist, and a second node that re-checks a result also receives a copy.

Questions or deletion requests: contact support from your account page. This page describes the product's behavior. Your Terms of Service and Privacy Policy govern.

Back to sign-up