AI, Model Router and Agent Terms

Version: 12 November 2026 (revision 2).

These terms apply when you use artificial intelligence, Model Router, agents, inference, AI search, AI-generated analysis or related automation supplied by Performance Hub Inc ("Performance Hub", "we", "us"). They form part of the Customer Terms. Capitalised terms not defined here have the meanings in the Customer Terms.

The Data Processing Addendum governs our processing of personal data on your behalf. The Surveillance, Audio and Biometric Terms additionally apply to those features. A completed Enterprise Deployment Schedule records any agreed enterprise configuration. The document-priority rules in the Customer Terms apply; this schedule does not override mandatory transfer terms or non-excludable rights.

1. AI services and terminology

Inputs are information submitted, connected, captured, retrieved or made available for AI processing by you, your authorised users or your configured services. They include prompts, instructions, files, images, recordings, audio, transcripts, facial or other metadata, search results, connected-app information, tool responses and context automatically assembled by an agent.

Outputs are information produced through AI processing, including text, code, transcripts, classifications, identifications, summaries, predictions and generated media. Actions are operations performed through an agent or automation, including interacting with software, connected accounts, APIs, files or physical devices. Inputs and Outputs are Customer Data to the extent provided in the Customer Terms; that classification does not confer ownership of another person's personal information or intellectual property.

AI services may use third-party cloud models, third-party models hosted on dedicated or local infrastructure, models operated by Performance Hub, or customer-supplied models where supported. A model's developer, its hosting operator and its execution location can be different. Third-party model limitations do not reclassify our own hosting, routing or other contracted services as someone else's services.

2. Your content, authority and use

You are responsible for determining whether your proposed use is lawful and appropriate, and for obtaining the rights, authority, notices, consents and other lawful bases needed to collect, connect, submit and process Inputs for that use. This includes information concerning staff, members, customers, visitors and other people, confidential third-party material, and content retrieved from connected accounts.

Your responsibility includes context submitted automatically under your configuration, not only messages that a person types. Before enabling access to a source, model or integration, assess what information it may make available and the authorised recipients and purposes. Do not enable a capability where you cannot meet the applicable requirements.

You must comply with the Acceptable Use Policy, applicable model restrictions and the law. You remain responsible for the business decisions, communications, publications and other uses you make of Outputs, subject to Performance Hub's own obligations under the Agreement and applicable law.

3. Models, routing and enterprise inference

On-premise AI and agent deployments are available only where included in an enterprise Order. An ordinary cloud subscription does not include an on-premise deployment.

Your enabled models and applicable routing configuration determine the inference destinations available to a request. Automatic routing may select among enabled models under those settings. A fallback is available only where the relevant model is enabled and the routing configuration permits its use. The presence of a model in the wider catalogue does not authorise routing contrary to a configured restriction.

Where only on-premise inference models are enabled, those are the available inference destinations; failure or insufficient capacity of a local model does not authorise a fallback to a disabled cloud model. For an on-premise agent using inference on the same local network, that local inference request path remains on that network. If a cloud agent uses on-premise inference, request content travels from that agent's cloud region to the on-premise inference location and the response returns to the cloud agent.

These descriptions concern the selected inference path. Separately configured backups, connected apps, external tools, data sources, support and management functions may involve other processing or transmission. They must be addressed in the applicable configuration and processing information. An on-premise model does not make an independently selected cloud agent, cloud backup or external integration local.

An enterprise local-only, hybrid or other agreed boundary is recorded in the Order or Enterprise Deployment Schedule. Your authorised administrators are responsible for selecting appropriate models and settings and controlling who can change them. We remain responsible for supplying the contracted routing controls and respecting applicable instructions and processing restrictions.

4. Agents, permissions and supervision

You decide which sources, capabilities and connected services to make available to an agent, and the instructions, schedules and approval settings under which it operates. Configure access appropriate to the work, review permissions when responsibilities change, protect credentials, and monitor your ongoing automations.

An agent may assemble context, make several model or tool calls, and perform Actions without a person reviewing each intermediate step where your settings permit that operation. A general instruction may produce multiple operations. You must assess the potential effects before authorising access or unattended operation and use appropriate approval requirements, testing and fallback arrangements for consequential Actions.

AI-generated content or external material may contain inaccurate or malicious instructions. Filters, redaction, content controls and prompt-injection protections can assist with identified risks but do not certify lawful use or detect every problem. Check which controls are enabled and whether a key, agent or connected service overrides them. The availability of a control is not a representation that it is enabled for every request.

You must not grant access that you are not entitled to grant. Your responsibilities do not excuse our failure to apply the access controls or other protections we are required to provide.

5. Accuracy, limitations and reliance

AI is probabilistic. Outputs may be false, incomplete, outdated, biased, misleading, offensive, inconsistent or unsuitable, and similar Inputs may produce different Outputs. Search may omit relevant material. Transcripts, speaker labels, facial matches and other classifications can be wrong. Generated software or instructions can contain defects or have unintended effects.

You are responsible for evaluating Outputs and Actions for your use, including independent verification and appropriate human review before publication, reliance or a consequential decision. Outputs and Actions are not legal, medical, safety, security, employment, financial or other professional advice, and are not a certification of identity, legal compliance, evidential completeness or the absence of a threat. You must not treat a model's confidence score or fluent explanation as proof of correctness.

Agents, Model Router and Outputs are not supplied as certified life-safety systems or staffed monitoring or emergency-dispatch services unless the applicable capability is expressly included in an Order and any required certification is identified. Do not rely on an Agent, model output, internet connection or hosted notification as the only means of emergency response, emergency access or safe egress. You may use supported access-control and incident workflows for their agreed ordinary purposes while maintaining the required independent emergency safeguards. The safety limitations in the Customer Terms apply.

Except for an express commitment in an applicable Order or a non-excludable legal requirement, we do not warrant a particular result, accuracy level, completeness, uniqueness, fitness for a particular AI use, or that Outputs will be free of third-party claims. These limitations do not remove our express service obligations or liability that cannot lawfully be excluded or limited.

For applicable AI, automated-decision and product laws, you are responsible for the deployer, operator or equivalent obligations that apply to your own deployment, including notices, human oversight, prohibited-practice restrictions and impact assessments that the law places on you. These terms do not transfer those duties to Performance Hub or classify every workload.

6. Third-party models and services

You must comply with the applicable third-party model licences and usage restrictions identified in the model information or Order made available for your selected service. Those restrictions may concern permitted use, distribution, attribution, supported territories and prohibited activities. No permission under the Agreement gives you rights that a relevant rights holder has not granted.

We may restrict or withdraw access where a provider or rights holder requires it or where continued provision would breach applicable law or contractual restrictions. Changes, notice, suspension and any resulting remedies are subject to the Customer Terms and applicable law. A reference to provider terms does not permit us to impose undisclosed charges or override an agreed data-processing boundary.

A third party may change a model, impose rate or capacity limits, refuse a request, suspend availability or discontinue features. We do not independently warrant or guarantee that party's model quality, availability, interoperability, response times or continuing supply, and that party is not a party to the Agreement. Where we engage a provider to process personal data on your behalf, the applicable DPA and subprocessor obligations continue to apply. A third-party designation is not a blanket exclusion of those duties.

Where a supported integration uses an account you contract for directly with another provider, you are responsible for that account and its terms. A Performance Hub API key used by an external application is still a key to Performance Hub; it does not by itself establish a separate contract between you and an upstream model provider.

7. Availability, performance and support

Model performance depends on the workload, model, configuration, hardware, concurrent demand, context size, connectivity and other dependencies. Published examples, benchmarks, estimates and model descriptions do not create a service-level commitment unless expressly incorporated as such in an Order.

No uptime, latency, throughput, error-rate, recovery-time or support-response SLA applies to these features unless expressly agreed in writing. An upstream provider's SLA or service-credit programme does not automatically apply between you and Performance Hub. Credits or other remedies apply only where the Agreement or applicable law provides for them.

Local inference may depend on licensed software, sufficient capacity, maintained hardware and the connectivity required by the agreed deployment. On-premise supply does not imply indefinite support, a perpetual managed service, every future model, or unlimited capacity. Any purchased support period or special commitment must be identified in the Order, without reducing legally required rights or support.

Raise support requests through Performance Hub Support. Performance Hub staff access for remote support must be initiated and authorised by you. That authorisation is limited to the support purpose and scope you permit. Automated operation and management already included in the agreed configuration are distinct from staff access; requesting support does not authorise unrelated use of Customer Data.

8. Metering, budgets and payment

AI features may incur charges for model requests, tokens, audio, images, compute, storage, tool use or other published units, as identified in the applicable price information or Order. An agent task may involve multiple calls and operations. A request's business usefulness is not itself the billing unit.

The pricing sources, priority and variable-pricing rules in sections 6 and 7 of the Customer Terms apply. Review the current Model Router catalogue and Billing & Cost Management → Pricing, subject to any agreed Order or custom rate card. Model rates may change frequently, including automatically following upstream provider changes, without separate individual notice unless required by applicable law or your Order. Changes apply prospectively under those sections and do not override an agreed fixed rate or pricing formula.

Automatic routing, retries, fallbacks and separate calls within a continuing agent task may use different enabled models with different applicable rates. Each billable call is charged at the rate applicable to that call under the Customer Terms. A previous request, estimate or prepaid credit purchase does not lock the price of later calls unless expressly agreed. You are responsible for reviewing rates and managing your enabled models, routing settings, scheduled tasks and authorised ongoing usage.

Retries, fallbacks, cached operations, refused requests or interrupted tasks are chargeable only to the extent the disclosed charging rules and actual metered usage provide. Your payment, cancellation and refund rights are governed by the Customer Terms and any applicable Order.

You are responsible for the settings and charges authorised by your users, including auto-top-ups and changes to permitted spending. Review the scope of each budget or limit and distinguish a usage estimate or reporting delay from an enforced spending control. This clause does not redefine a control described or agreed as a hard limit as an estimate, or authorise charges above a limit we have agreed to enforce.

Local inference pricing and any software, hardware or management charges are those in the enterprise Order. Local inference does not automatically incur cloud-model charges merely because it uses the Model Router interface.

9. Data handling, retention and training

Data handling depends on the selected deployment and features. Model processing, agent conversation history, workspace files, search indexes, memory, prompt caches, logs, backups and provider security or abuse-monitoring records are distinct activities. A statement about one component's logging does not establish that every other component retains no content.

Our handling of personal data is described in the Privacy Policy and, for processing on your behalf, the DPA and applicable schedules. An enterprise configuration identifies relevant permitted transfers, locations and external services. We do not acquire a general right under this schedule to use Customer Data for independent model training. Any additional use requires the authorisation and legal basis required by the Agreement and applicable law.

Claims about provider training, retention or processing location must be read with the terms and configuration applicable to the selected service. These terms do not promise universal zero retention, no provider access or a single processing country across every selectable model and integration.

You are responsible for lawful retention choices and for independent exports or copies you control. Removal of a source recording, file, model or integration may involve separate treatment of transcripts, indexes, agent memory and backups. Applicable return, deletion and assistance obligations remain governed by the DPA and the documented lifecycle, including lawful preservation requirements.

10. Intellectual property and customer-supplied models

You retain the rights you already hold in Inputs. Subject to third-party rights and applicable licences, we pass through to you any rights in Outputs that we are entitled to pass through under the Agreement. We do not promise that an Output is protectable, exclusive or free from material in which others hold rights.

The processing licence in the Customer Terms permits the operations necessary to provide the selected Services, including authorised routing and tool use. It does not transfer ownership of your Inputs or create rights over individuals' personal information beyond lawful processing.

For customer-supplied models, software, weights, fine-tuning data or integrations, you must have the licences and permissions needed for the agreed hosting, processing and use. Ownership and permitted use of separately commissioned fine-tuned weights, adapters or other deliverables must be stated in the applicable Order; neither party acquires rights that the underlying model licence prohibits.

11. Liability and prohibited uses

The exclusions, liability limits, customer indemnity and non-excludable rights in the Customer Terms apply to this schedule. Claims relating to Inputs, configuration, instructions, Outputs, Actions or deployment decisions are covered by the customer indemnity only within its stated scope, responsibility allocation, financial cap and exceptions. This schedule does not create a separate or unlimited indemnity or an additional Performance Hub indemnity for Outputs or Actions.

You must not use AI features for a prohibited use under the AUP or applicable law. A general ability to run a model, a provider's availability in the catalogue or an Output suggesting a course of action does not authorise an unlawful activity or an unsupported regulated deployment.

12. Contact

Preferred contact: Submit a support ticket.

Performance Hub Inc
251 Little Falls Drive, Wilmington, DE 19808, United States
Telephone: +1 (302) 244-5288