YAKKL AI and Agentic Services Terms

AI systems are probabilistic and can produce Output that is inaccurate, incomplete, misleading, biased, offensive, outdated, insecure, non-unique, or fabricated. They may misunderstand instructions, omit constraints, cite nonexistent sources, disclose unintended information included in context, or behave differently across runs.

AI-generated code can compile and pass tests while still containing logic defects, security vulnerabilities, race conditions, data-loss paths, incompatible dependencies, or license problems. Plans and analysis can rely on false assumptions. Retrieval can miss or mis-rank relevant information. Tool calls can target the wrong resource.

YAKKL does not represent that an AI model reasons, understands, verifies truth, or exercises professional judgment. Model confidence, fluent language, a passing automated check, or an agent's statement that work is complete is not proof of correctness.

You are responsible for deciding whether and how to use Output and Actions. Before using, merging, deploying, publishing, executing, or relying on them, apply review proportionate to the risk, including as appropriate:

• inspect source changes, diffs, commands, plans, citations, dependencies, and generated artifacts;
• run relevant builds, tests, linters, security scans, license checks, and manual validation;
• confirm facts against authoritative and current sources;
• maintain backups, version control, recovery paths, access controls, and deployment approvals;
• confirm that requested Actions stayed within scope and did not create unintended external effects; and
• obtain qualified professional review for consequential legal, financial, tax, medical, employment, privacy, security, safety, or regulatory matters.

AI Output is not professional advice and must not be the sole basis for a decision with legal or similarly significant effects on a person.

"Actions" are operations initiated or proposed through the AI Services, including reading, creating, altering, moving, or deleting files; running shell commands or code; installing software; accessing networks; using credentials; changing repositories; opening pull requests; deploying software; interacting with infrastructure, databases, calendars, communications, or third-party services; purchasing or consuming resources; or publishing content.

Actions can have external, costly, security-sensitive, or irreversible effects. You control the permissions, tools, accounts, approval gates, and environments made available to an agent. An agent's technical ability to perform an Action does not establish that the Action is authorized.

You must use approval controls appropriate to the risk. Do not provide an agent broader credentials or access than needed. Review consequential Actions before execution unless your authorized policy intentionally permits unattended execution and includes monitoring, limits, and recovery controls.

YAKKL may provide guardrails, confirmations, plan enforcement, audit trails, or recovery features, but they do not eliminate your responsibility for configuration and review. YAKKL may block or require confirmation for an Action believed to present security, legal, abuse, or material financial risk.

These AI Terms allocate the inherent risks of probabilistic AI systems. They do not disclaim YAKKL's responsibility under the Base Terms for a material failure of a YAKKL-provided permission-enforcement, approval-gate, or other documented safety mechanism to function materially as documented; any such claim remains subject to the Base Terms' warranties, disclaimers, and limitations of liability.

Consumption of CU, usage-based Services, or third-party resources by an agent, automation, or workflow operating within the permissions, limits, and configurations you or your administrators enabled is authorized usage for billing purposes, as further described in the Subscription and Credit Unit Terms.

Adversarial content and prompt injection. Content an agent encounters — including repositories, dependencies, web pages, documents, and tool or integration results — may contain instructions crafted by third parties to mislead an AI system into unintended Actions, including disclosure of data or destructive operations. Treat external content as untrusted; use the isolation, scoping, and approval controls made available; and do not give an agent broader credentials or network access than the task requires. YAKKL provides safeguards designed to reduce these risks but cannot guarantee detection of adversarial content. An instruction injected by a third party is not an instruction from you, and Actions it induces are not authorized by YAKKL; responsibility for resulting loss is allocated under the Base Terms.

You retain rights in Customer Content and own Output as stated in the Base Terms. Output may be the same as or similar to output provided to others. Ownership does not guarantee novelty, copyrightability, patentability, non-infringement, or freedom to use.

YAKKL does not use Customer Content or Output to train AI models and does not authorize model providers or other third parties to train on it on YAKKL's behalf. YAKKL processes Content only as needed to provide, secure, support, and administer the selected Services, follow instructions, prevent abuse, and comply with law.

YAKKL personnel do not inspect Content as part of ordinary analytics. Personnel access is limited to support you request, security or abuse investigation, legal compliance, or operation of a managed feature that cannot be provided through automated processing alone. Access must be authorized and appropriately logged or recorded.

If you intentionally send Content directly to a provider or custom endpoint through BYOK, that provider or endpoint handles it under the terms and settings you select. YAKKL cannot alter or guarantee an independent provider's practices.

Output may include or resemble third-party code, data, text, packages, APIs, or other material. Some code or dependencies may be subject to open-source or other licenses. Provider filters, citations, provenance indicators, similarity checks, or license scanners may be incomplete.

You are responsible for reviewing provenance, applicable licenses, attribution, notices, export restrictions, security, and intellectual-property risk before using or distributing Output. Do not remove license or attribution notices without authority. YAKKL does not grant rights in third-party material.

YAKKL may support these modes:

1. Direct BYOK. A customer-controlled client sends a request using the customer's credentials directly to the selected provider. Request and response payloads — prompts, Content, and completions — travel from the client to the AI provider associated with your API keys; YAKKL does not receive those payloads merely because its client initiates the connection. Separately from request payloads, the client may send operational telemetry, diagnostics, and usage data to YAKKL as described in the Privacy Policy; that data is designed to exclude Content. Unless you select another supported route, BYOK defaults to this direct path. BYOK credentials must be API credentials that the provider's terms permit you to use with third-party tools; you are responsible for the eligibility of the credentials you supply.
2. Custom endpoint. The customer sends a request to an endpoint it chooses, including its own backend. The endpoint's operator controls processing, logging, terms, and charges.
3. YAKKL-managed routing. YAKKL receives and routes a request to one or more providers for execution, optimization, retrieval, safety, or reliability. This path may consume CU or incur a fee and is governed by YAKKL's subprocessor commitments.
4. Local execution. Supported models or tools operate on the customer-controlled device or environment. Content remains local unless another enabled feature transmits it.
5. Hybrid execution. A workflow uses more than one mode. The Service will identify the mode or applicable policy at a level reasonably sufficient for customer control and reconciliation, and any step that uses YAKKL-managed processing is governed by the managed-routing commitments regardless of how the workflow is labeled.

Customers may change routing choices, subject to plan, model, provider, security, and technical constraints. Provider availability, model versions, context limits, safety systems, pricing, and terms change. YAKKL may substitute managed providers or models based on capability, quality, security, compliance, availability, latency, or cost. Where practicable, YAKKL will disclose the effective model/provider in usage records or documentation, and will use commercially reasonable efforts to provide advance notice before removing a managed model or provider in a way that materially affects documented workflows.

Customer-selected providers, custom endpoints, integrations, and BYOK accounts are governed by their own terms and privacy practices. You are responsible for reviewing and complying with them. YAKKL does not control their retention, safety classification, model training, service changes, or charges.

For YAKKL-managed routing, YAKKL will contract with providers as subprocessors where applicable and maintain a current register. YAKKL's agreements with managed-routing model providers require that they not use Customer Content or Output to train their models. Providers may apply automated abuse or safety classifiers. Content flagged by those systems or YAKKL safeguards may be retained and reviewed by authorized personnel for the limited time reasonably necessary to investigate, secure the Services, enforce policy, or comply with law. YAKKL will not treat abuse review as permission to train models.

AI Services may create conversations, plans, indexes, embeddings, caches, artifacts, audit trails, tool records, or other Customer Content. If stored only locally, that Content remains under the customer's control. If YAKKL cloud storage or logging is enabled, retention follows the applicable plan, workspace settings, order, and Privacy Policy or DPA.

You may delete eligible cloud Content through available controls. Deletion may not immediately remove limited encrypted backups, billing records, security records, legal holds, or abuse-review records. Indexes, embeddings, or caches derived from deleted Content will be deleted or rendered unusable according to the applicable retention process.

YAKKL may retain de-identified Usage Data that excludes Content for analytics, model-selection statistics, feature usage, traffic, reliability, and billing.

YAKKL-managed execution may consume CU. CU are blended units of service that may account for model input and output, reasoning, cache behavior, retrieval, routing, tool execution, compute, storage, and other managed work. They are not raw model tokens and do not represent a fixed amount of vendor cost or compute.

YAKKL will provide good-faith usage information in the Service, including available balances and, where feasible, breakdowns by model, workflow, or operation. Provider measurement, reporting delays, cache adjustments, retries, model behavior, and blended pricing may prevent exact real-time attribution. Usage displays are for practical reconciliation, not a guarantee of a specific token count, vendor rate, margin, or cost component. The Subscription and Credit Unit Terms control.

You may not use the AI Services in a way prohibited by the AUP or applicable provider policies. Uses involving critical infrastructure, weapons, safety systems, employment, housing, credit, insurance, healthcare, education admissions, legal services, biometric identification, or other high-impact decisions require lawful authority, qualified human review, testing, monitoring, and any controls specified by YAKKL.

The AI Services must not be used as the sole decision-maker for a decision producing legal or similarly significant effects on a person.

Do not place credentials, sensitive signing material, authentication tokens, production secrets, regulated data, or other sensitive information into prompts or tools unless necessary, authorized, and protected by appropriate controls. Use secret-management features, least privilege, scoped credentials, and test environments when available.

You are responsible for credentials supplied through BYOK or custom endpoints. YAKKL will use reasonable safeguards for credentials stored in YAKKL-managed systems, but cannot protect credentials exposed through customer code, logs, prompts, repositories, tools, or third-party endpoints.

Preview, beta, research, and evaluation AI Services may be less reliable, may lack complete security or compliance features, may change without notice, and may not be suitable for production or regulated data. Unless an Order expressly states otherwise, no service level, support commitment, or continued availability applies to those features.

YAKKL may limit, suspend, or block a model, provider, tool, Action, or workflow to address abuse, security, provider restrictions, legal obligations, availability, or material cost anomalies. When practicable, YAKKL will explain the restriction and provide an appeal path.

YAKKL may update these AI Terms as technology, law, providers, and Services evolve. Material adverse changes to an existing paid Service will receive notice under the Base Terms.

The Base Terms' disclaimers, liability limits, indemnities, dispute provisions, and termination rules apply to the AI Services. If these AI Terms conflict with the Base Terms on an AI-specific issue, these AI Terms control. Mandatory law is not waived.