Abacus.AI does not train models on your data, but enterprise chat records still need their own retention limits
When evaluating Abacus.AI for team use, check model training, platform storage and administrator access separately. Retention settings for individual deployments can also override enterprise policies.
Teams preparing to connect an internal knowledge base to Abacus. AI should not end their review at “not used for training.” For ChatLLM and custom chatbots, the more practical questions are how long records remain on the platform, who can see them, and whether the settings actually apply to this deployment. This article is based on official documentation; it is not a security audit or a compliance determination.
Not training and not storing are two different commitments
ChatLLM’s data security FAQ states that Abacus. AI does not use user data for training and has enterprise agreements with external model providers to protect customer data from being used for training. However, the developer platform’s access management documentation also states that chatbot data is retained indefinitely on the Abacus platform. External model providers’ zero-day retention policies are a separate arrangement, and enterprise customers can adjust retention policies for chatbots on the platform.
Therefore, if a company requires discussion records to be deleted when a project ends, a provider’s commitment not to train models does not meet that requirement. Confirm the actual retention settings for the product, contract and workspace in use with procurement or an administrator. Do not interpret zero retention on the external model side as meaning that the Abacus platform does not store records either.
An enterprise policy may not cover every deployment
The official Setting and Permissions documentation identifies these governance capabilities as enterprise features managed by organization administrators. Organization Retention Policy sets the retention period for deployment conversations in hours, with automatic deletion once that period expires. Retention settings for a specific deployment override organization-level settings.
For example, a team that wants conversations retained for only seven days can use 168 hours as its intended configuration. But acceptance checks should go beyond a screenshot of the organization settings page: check whether the actual deployment has a different retention period, and verify the expected behavior using test content that contains no real confidential information. This is a recommended checking method, not a deletion outcome verified in this article. Ordinary paid accounts should not be assumed to include these enterprise controls either.
Private by default still has an administrator boundary
The same enterprise administration documentation explains that, when private-by-default is enabled, new conversations are visible only to their creators, and users can explicitly share them with specified users or groups. Organization administrators can still access all conversations. Disabling public conversation sharing also invalidates existing public links. These settings address internal visibility and external links respectively; neither can replace the other.
A pilot is appropriate when the team can clearly explain retention periods, deployment exceptions, the scope of administrator access and sharing rules, and someone is responsible for subsequent checks. If the requirement is that no administrator can view the content, or if the platform’s default long-term retention is unacceptable, resolve access and retention conditions before connecting real internal materials. The fact that a product offers controls does not mean your workspace is already configured to meet your company’s requirements.
Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.
Original source:abacus.ai