Terms for AI Features and Automated Actions

"AI Features" means the features of the Service that use machine-learning or generative AI models, including drafting ActionBooks and Functions, suggesting an ActionBook for an event, and answering questions about Customer Data. "Input" means Customer Data or other content that you or the Service submits to an AI Feature, and "Output" means the content an AI Feature returns.

2.1 WhaTap may run AI Features using models hosted by third-party AI providers, which act as WhaTap's subprocessors and are listed at virca.ai/subprocessors. An AI provider may process Input outside the United States, as described in that list. WhaTap will give notice of a new AI provider as described in the Data Processing Addendum.

2.2 WhaTap does not currently apply automated removal of personal data or secrets to Input before it reaches an AI provider. Configure the Worker's collection so that Customer Data does not contain information you do not want processed by AI Features.

WhaTap does not use Customer Data, Input, or Output to train or fine-tune AI models. WhaTap does not authorize its AI providers to use Input or Output to train models, and uses AI providers that state they do not log or train on this content. This covers personal data within Customer Data. WhaTap will not begin using your Customer Data, Input, or Output for training or fine-tuning without first giving you at least 30 days' written notice and obtaining your written consent, and will not use data collected before your consent. It does not restrict WhaTap's use of Usage Data or of aggregated, de-identified data as permitted by Section 8.3 of the Agreement.

Input and Output are Customer Data. As between the parties, you own Output, excluding WhaTap's pre-existing materials such as Builtin Functions, templates, and the Service's prompts and system instructions. Output may be similar to output generated for other customers.

5.1 Output is generated automatically and may be incomplete, inaccurate, or unsuitable for your environment. You are responsible for reviewing Output before you rely on it, and in particular before you save an AI-drafted Function or ActionBook or bind it to a rule. WhaTap does not warrant that Output is accurate, complete, or current.

5.2 An AI-drafted Function or ActionBook becomes a Customer Function or an ActionBook you created once you save it, and Part B applies to it in the same way as to one you wrote yourself.

6.1 You may connect AI Clients to the Service through its MCP interface. An AI Client runs under your account and your agreement with its provider, not WhaTap's. Data the Service returns to an AI Client leaves the Service at your direction, and WhaTap is not responsible for how the AI Client or its provider processes it. WhaTap does not control the AI Client, the model behind it, who can use it, or the content it acts on, including instructions embedded in data it reads.

6.2 An AI Client can request Actions. Requests from an AI Client are subject to the Severity rules for AI-originated requests described in the Documentation and to your Approval Configuration. The Service is designed to refuse critical-Severity Actions requested by an AI Client, but you must not rely on that control as your only safeguard. You are responsible for which AI Clients you authorize and for the scope of the credentials you issue to them. Treat every credential issued to an AI Client as able to request Actions.

6.3 The Audit Log records that an Action was requested through the MCP interface. It may not identify the specific AI Client or model that made the request, so keep your own records if you need that attribution.

Each party will comply with the laws on AI that apply to it. If you pass Output to others, you are responsible for any disclosure the law requires of you.

You will not (a) use AI Features to develop or train a model or service that competes with the Service; (b) try to extract the Service's prompts, system instructions, or model parameters, or circumvent safety controls; or (c) submit Input that you do not have the right to submit.

9.1 Closed execution. The Worker runs Actions only as ActionBooks that call Functions registered in your workspace. An ActionBook cannot run commands other than through those Functions.

9.2 Your Approval Configuration. You choose, for your workspace, whether Actions that match a rule run without a further approval or wait for approval, and you choose which ActionBooks to bind to which rules. WhaTap does not choose this setting for you. If you require approval, every Action that rule triggers waits for approval, whatever its Severity. Binding an ActionBook to a rule in a mode that does not require further approval is your advance approval of every Action that rule triggers, including Actions graded critical.

9.3 Approvals. An approval given by an Authorized User in the Service or in a connected team-chat channel is your instruction to run the Action shown. An approval request that is not answered within the timeout stated in the Documentation is treated as rejected. The Service does not currently restrict which Authorized User may approve an Action, including the Authorized User who requested it, and anyone who can reach a team-chat channel you configured for approvals can approve an Action shown there. If you need approvals limited to particular people, limit workspace membership and channel access accordingly.

9.4 Builtin Functions. Builtin Functions and ActionBooks supplied by WhaTap are provided disabled, and run in your workspace only after you enable them. Review each one before enabling it. WhaTap assigns their Severity based on its assessment of whether the command's effect can be reversed; that assessment is guidance, not a guarantee that a Function is safe in your environment.

9.5 Customer Functions. You assign the Severity of each Customer Function. The Service treats a Customer Function with no declared Severity as normal. The approval screen shows each Function's signature and description, not its underlying command body, so review Customer Functions when you save them.

10.1 Privileges. The Worker runs Actions with the privileges of the user it runs as. On Linux Servers, that is root by default, or another user you choose. Some Functions may need root privileges.

10.2 Running Actions. Once an Action has started on a Server, it may run until it finishes or reaches its timeout, and the Service may not be able to stop it partway through. Targets that have not yet started can be cancelled.

10.3 Outcomes. The Service runs the steps of an ActionBook and records what the Worker reports. It does not independently determine whether an Action or a rollback step achieved its intended result. Reverse steps run only if the ActionBook's author wrote them. Evidence the Service collects for the record may be incomplete, and the Service does not determine root cause; conclusions drawn from that evidence are yours.

10.4 Connectivity. Actions do not run on a Server while its Worker is disconnected from the Service, and events and data that occur during a disconnection may not be collected. The Service detects only the conditions your rules cover; a Server that stops reporting altogether produces no data, and may never raise an event.

10.5 Audit Log. The Service keeps a system-maintained, append-only record of Actions, including who approved them. Very large results may be truncated in the record. The Service does not offer an Audit Log export. You can ask for a copy at ask@virca.ai.

You are responsible for: (a) your Approval Configuration and the rules you bind ActionBooks to; (b) the content and Severity of Customer Functions and ActionBooks you create or save, including AI-drafted ones; (c) reviewing Builtin Functions before enabling them; (d) deciding which Servers the Worker is installed on and which Actions may run on them; (e) testing ActionBooks in a non-production environment before using them in production; (f) maintaining backups and recovery procedures for your Servers; (g) who belongs to your workspace and can approve Actions; and (h) keeping whatever independent monitoring, alerting, and incident-response procedures you need to catch what the Service does not. The Service does not replace the judgment of your engineers, and you should not make it your only safeguard against failure, data loss, or a security incident.

12.1 Actions within your configuration. As between the parties, you are responsible for the results of an Action that the Service runs as permitted by your Approval Configuration and the Severity rules described in the Documentation, whether it was triggered by a rule, requested by an Authorized User, or requested by an AI Client. TO THE MAXIMUM EXTENT PERMITTED BY LAW, WHATAP IS NOT LIABLE FOR LOSS, DAMAGE, OR DISRUPTION ARISING FROM SUCH AN ACTION, EXCEPT TO THE EXTENT IT IS FINALLY DETERMINED TO HAVE RESULTED FROM WHATAP'S GROSS NEGLIGENCE OR WILLFUL MISCONDUCT.

12.2 Actions outside your configuration. WhaTap is responsible, subject to Section 14 of the Agreement (including the Early Access cap), for losses caused by the Service running an Action that your Approval Configuration and the Severity rules did not permit, or running an Action materially differently from the ActionBook and Functions approved, because of a defect in the Service.

12.3 No other warranty of results. Except as stated in Section 12.2, WhaTap does not warrant that any Action, ActionBook, or Builtin Function will resolve a problem or achieve a particular result.

You will not use Actions (a) against any system you do not own or are not authorized to operate; (b) to disrupt, damage, or gain unauthorized access to any system or data; or (c) in any system where the failure or misoperation of an Action could reasonably be expected to cause death, personal injury, or serious physical or environmental damage, unless WhaTap agrees in writing.

WhaTap may update these terms as described in Section 18 of the Agreement. An update will not, without your consent: (a) expand WhaTap's rights to use Customer Data, Input, or Output; (b) remove the commitment in Section 3; or (c) move an Action from requiring your approval to running without it.

AI Features and Actions provided as Early Access Services are subject to Section 5 of the Agreement. Sections 3 and 12.2 of these terms still apply to them.