AI Services Addendum

YAKKL Meridian™ and YAKKL Ari™ use large language models and other AI systems to generate code, plans, analysis, documentation, and other output. These systems have real limitations that you need to understand:

• They produce wrong answers. AI models hallucinate — they generate confident-sounding responses that are factually incorrect, logically flawed, or completely fabricated. This isn't rare; it's a known characteristic of how these systems work.
• Code may compile but still be broken. AI-generated code can pass syntax checks and even basic tests while containing subtle bugs, security vulnerabilities, race conditions, or logic errors that only surface in production.
• Plans and analysis can have blind spots. AI-generated architecture, plans, and analysis may miss critical constraints, make invalid assumptions about your infrastructure, or overlook edge cases that a human familiar with the system would catch.
• Information can be outdated. AI models are trained on data with a cutoff date. They may reference deprecated APIs, outdated best practices, removed features, or incorrect version information.
• Output varies between runs. The same prompt can produce different results on different occasions. Consistency is not guaranteed.

None of this is a defect — it's how the technology works today. We're upfront about it because you need to plan for it.

Every piece of AI-generated output is a draft, not a deliverable. You are responsible for reviewing, testing, and validating everything before it goes anywhere that matters. Specifically:

• Code: Run your test suite. Review for security issues. Check edge cases. Don't merge AI output without the same scrutiny you'd give a pull request from a new hire.
• Architecture and plans: Verify assumptions against your actual system. Check that dependencies exist and are compatible. Validate cost and performance estimates independently.
• Analysis and recommendations: Cross-reference against authoritative sources. Don't take AI-generated data points at face value.
• Anything you ship, deploy, publish, or rely on for business decisions: The AI wrote it; you own it. Treat it accordingly.

If AI-generated output causes issues in your systems, data loss, security incidents, incorrect business decisions, or any other harm — the responsibility sits with whoever approved and deployed it. That's you, not YAKKL.

YAKKL Meridian™ and YAKKL Ari™ can help with planning, drafting, research, analysis, and workflow execution across many domains. That does not make them qualified professionals.

AI output is not a substitute for:

• Legal advice from a licensed attorney
• Financial or investment advice from a qualified advisor
• Tax guidance from a CPA or tax professional
• Medical advice from a healthcare provider
• Security assessments from a qualified penetration tester or security engineer
• Any other domain where professional licensing, certification, or specialized training is required

If you need professional advice, hire a professional. The AI can help you prepare questions for that professional, organize your thinking, or draft documents for their review — but the professional makes the call, not the AI.

A YAKKL Meridian™ subscription gives you access to the features, usage limits, and capabilities described for your plan at the time of purchase. Here's what that means in practice:

• Plan names and features: What's included in each tier (Individual, Team, Enterprise) is shown at checkout and on our pricing page. We don't change what you're paying for mid-cycle without notice.
• Usage limits: Each plan has limits on things like context tokens, managed execution volume, and concurrent sessions. These limits exist because AI infrastructure costs real money per request.
• Seat counts: Team and Enterprise plans are licensed per seat. Adding seats mid-cycle is prorated; removing seats takes effect at next renewal.
• Founding preview pricing: If you subscribed during a founding preview, your promotional rate is locked for the stated duration. After that, standard pricing applies.

Current plan details and pricing are always available at meridian.yakkl.com/pricing.

Credit Units ("CU") are how we meter managed execution in YAKKL Meridian™. A few things to be clear about:

• CU are not money. They're not a stored cash balance, prepaid token, cryptocurrency, gift card, or anything you can transfer, sell, or redeem for cash. They're a metering unit.
• What consumes CU: Model inference (the biggest factor), context processing, tool execution, retrieval operations, compression, cache-aware workflows, and other managed execution activity. Different operations cost different amounts of CU depending on the model used, input/output size, and complexity.
• CU included with plans: Each subscription tier includes a monthly CU allocation. Unused CU don't roll over. If you run out, you can purchase additional CU blocks or wait for your next billing cycle.
• CU pricing: The per-unit price of additional CU blocks may vary by plan tier and purchase volume. Current pricing is shown in your dashboard.
• Cache credits: When prompt caching reduces actual provider costs, those savings are credited back as CU. This incentivizes efficient prompt design — you pay less when your workflows cache well.

We publish a CU cost schedule that maps common operations to approximate CU consumption. This schedule is a guide, not a guarantee — actual consumption depends on model behavior, input complexity, and execution path.

YAKKL Meridian™ supports two execution modes, and the billing and data handling differ significantly between them:

Bring Your Own Key (BYOK):

• You supply your own API keys for AI providers (Anthropic, OpenAI, Google, etc.)
• Requests go directly from Meridian to that provider using your key
• The provider bills you directly for model usage under their terms and pricing
• YAKKL does not see, store, or process the content of your prompts or completions beyond what's needed to route the request
• You're responsible for complying with the provider's acceptable use policies
• YAKKL is not liable for charges incurred on your provider account

Managed Execution (YAKKL-routed):

• YAKKL handles model selection, routing, and provider management on your behalf
• Execution consumes CU from your plan allocation
• YAKKL may route requests to different providers or models based on cost, availability, capability, and your plan tier
• You don't need to manage API keys, provider accounts, or billing relationships with AI vendors

You can use both modes in the same workspace. The choice is per-configuration, not per-account.

Paid YAKKL Meridian™ subscriptions work like this:

• Billing interval: Monthly or annual, selected at checkout. Annual plans include a discount. You can switch intervals — upgrading to annual takes effect immediately with prorated credit; downgrading to monthly takes effect at next renewal.
• Auto-renewal: Subscriptions renew automatically at the end of each billing period. We send a reminder email before renewal. If you don't want to renew, cancel at least 24 hours before your renewal date.
• Payment methods: Credit card or cryptocurrency, depending on the service. Crypto payments require sufficient balance including gas fees at the time of renewal.
• Failed payments: If a renewal payment fails, we retry for up to 7 days (credit card) or 3 days (crypto). During this period your service continues. After the grace period, your account downgrades to the free tier until payment is resolved.
• Taxes: Prices shown exclude sales tax, VAT, and similar taxes. These are added at checkout based on your billing address.
• CU add-ons: Additional CU blocks can be purchased at any time and are charged immediately. CU purchases are non-refundable.
• Seat changes: Adding seats is prorated for the remainder of your billing period. Removing seats takes effect at the next renewal — no partial refunds.

Cancellation stops future charges. Your access continues through the end of the paid period. Refunds are not provided except where required by law.

We run preview programs — like the Founding Builder Preview — to give early users access to YAKKL Meridian™ before general availability. If you're in a preview program, here's what to expect:

• Pricing may be promotional. Founding Builder pricing and bonus CU allocations are time-limited offers. The specific terms (duration, rate, CU bonus) are stated at enrollment. When the promotional period ends, standard pricing applies.
• Features are in development. Preview features may be incomplete, unstable, or significantly different from the eventual GA release. We'll try to minimize disruption, but things will change.
• Features may be removed. Something available in preview may not ship in GA, or may ship in a substantially different form. We won't charge you for features we remove, but we also can't guarantee that a specific feature will exist forever.
• Feedback shapes the product. Preview participants often have direct channels for feedback. We take it seriously — preview feedback has materially changed product direction in the past.
• No SLA during preview. We aim for high availability but don't make formal uptime guarantees during preview periods. If uptime guarantees matter for your use case, wait for GA or ask about an enterprise agreement.

How we handle the data that flows through YAKKL Meridian™ and YAKKL Ari™:

• We don't train on your prompts or outputs. Your code, plans, documents, and other content processed through Meridian are not used to train YAKKL's AI models or improve our products. Period.
• Session data: We retain session metadata (timestamps, token counts, model used, CU consumed) for billing, debugging, and service quality. This metadata does not include the content of your prompts or completions.
• Crash and error diagnostics: If an error occurs during execution, we may log diagnostic information to fix the issue. This may briefly include fragments of the request context. Diagnostic logs are retained for no more than 30 days and are not used for training.
• Third-party providers (BYOK): When you use BYOK, your prompts and completions are sent to the third-party provider. Their data handling policies, not ours, govern that data. Read their terms before providing your keys.
• Third-party providers (managed): When using managed execution, YAKKL routes your requests through our accounts with AI providers. We contractually require that these providers don't train on customer data, but their processing is governed by their terms in addition to ours.

AI is a fast-moving field. The models and providers available through YAKKL Meridian™ will change over time:

• We may add new models and providers as they become available.
• We may retire models that are superseded, deprecated by their provider, or no longer cost-effective.
• We may change default model routing to improve quality, reduce costs, or respond to provider changes.
• We may adjust CU consumption rates as provider pricing changes.

We'll give reasonable notice before making changes that materially affect your workflow. But "reasonable notice" for a model deprecation that a provider announces with 30 days lead time might be less notice than you'd like. That's the nature of building on a stack where the underlying providers are also evolving rapidly.

If a specific model version is critical to your workflow, pin it in your configuration. Pinned models are supported as long as they remain available from the provider, but we can't guarantee indefinite availability of any specific model.

Don't use YAKKL Meridian™ or YAKKL Ari™ to:

• Generate malware, exploits, or code designed to harm systems or users
• Produce content that violates law, infringes rights, or facilitates illegal activity
• Circumvent safety controls, content filters, or usage limits on AI providers
• Impersonate individuals or organizations
• Generate spam, phishing content, or social engineering material
• Conduct unauthorized security testing against systems you don't own
• Abuse provider APIs in ways that violate their terms or degrade service for others
• Build products that compete with YAKKL Meridian™ using output from YAKKL Meridian™

We reserve the right to suspend or terminate accounts that violate these rules. Egregious violations (malware generation, fraud facilitation) result in immediate termination without refund.

YAKKL Meridian™ depends on third-party AI providers, cloud infrastructure, and network connectivity. We don't control all of these, and outages happen:

• Provider outages may make specific models temporarily unavailable. Managed execution will attempt to route around outages where possible.
• We may take the service offline for maintenance with reasonable advance notice.
• Emergency maintenance (security patches, critical bug fixes) may happen without advance notice.
• We don't guarantee specific uptime percentages outside of enterprise SLAs.

If an outage materially prevents you from using the service for an extended period, contact support about billing adjustments.

Questions about YAKKL Meridian™ or YAKKL Ari™ terms, enterprise agreements, or billing: enterprise@meridian.yakkl.com

General support: support@yakkl.com