What AI Usage Markup Actually Costs You

When an AI platform proxies your model calls, the sequence is: your workflow sends a request to the platform, the platform forwards it to the model provider (OpenAI, Anthropic, Google, or another), the provider returns a response, and the platform forwards that response to you. At billing time, the platform charges you for the tokens used, plus a markup. The markup is typically expressed as a percentage of the underlying provider cost, a flat per-token surcharge, or both.

The mechanism matters because it means the cost is not fixed. It is a function of your usage volume. A 20% markup on $100 of monthly AI spend is $20. The same markup on $5,000 of monthly AI spend is $1,000. The markup rate does not change as you scale. Your bill does.

This structure creates a specific problem for teams that are succeeding with AI automation. The more value you extract from your workflows, the more tokens you consume, and the more markup you pay. Growth in AI productivity translates directly into growth in platform overhead. That is not a bug in the pricing model from the platform's perspective: it is the business model. But it is worth understanding clearly before you build production workflows on top of it.

There is a second, less visible cost: opacity. When a platform proxies your model calls, you lose direct visibility into what the provider is charging versus what the platform is charging. You see one bill, not two. That makes it harder to optimize, harder to audit, and harder to compare the true cost of different models or providers.

AI usage costs are driven by token volume, and token volume scales with three things: the number of workflow runs, the complexity of the prompts and context windows those runs use, and the number of users or automated triggers generating those runs.

Workflow runs multiply token consumption

A single workflow run that calls an AI agent might consume a few thousand tokens. A team running hundreds of workflow instances per day consumes millions. At each of those tokens, a markup-based platform collects its percentage. The math is straightforward: markup cost equals token volume multiplied by the per-token markup rate. As your team automates more, token volume grows, and so does the markup line.

Richer context windows cost more

Modern AI workflows often pass substantial context to the model: a document to summarize, a conversation history, a set of retrieved records, a system prompt with detailed instructions. Larger context windows mean more input tokens per call. For teams doing document processing, content generation, or multi-turn agent interactions, the per-call token count can be significantly higher than a simple prompt-and-response exchange. Markup compounds on every one of those tokens.

Multiple providers mean multiple markup surfaces

Teams running production AI workflows rarely use a single model for everything. A content team might use one model for drafting and another for classification. A finance team might use a fast, cheap model for extraction and a more capable model for analysis. Each provider relationship is a separate markup surface on a proxying platform. On a BYO-keys platform, each provider relationship is a direct billing relationship with the provider at their published rate.

The compounding effect over time

Markup costs are not a one-time expense. They recur every billing cycle, at the same percentage of an ever-growing usage base. A team that doubles its AI workflow volume over twelve months doubles its markup spend over the same period, with no corresponding increase in the value the platform provides for that incremental cost. The markup is not buying you more capability as you scale. It is buying you the same proxy service on a larger volume of calls.

DAVE connects to AI providers through a Provider record you configure in the platform. Navigate to Admin, then API Management, then Providers. Each Provider record stores the service type, your API key, the base URI for the service, and a catalog of the models you want to make available to your agents.

DAVE supports seven provider types: OpenAI, Anthropic, Azure OpenAI, Google AI, OpenAI-Compatible, Local (Ollama), and Custom. For each type except Custom, DAVE pre-fills the default base URI for the service (for example, https://api.openai.com/v1 for OpenAI, https://api.anthropic.com/v1 for Anthropic, https://generativelanguage.googleapis.com/v1beta for Google AI). You supply your API key. DAVE stores it encrypted using AES-256-GCM authenticated encryption and never returns it through the API once stored.

When a workflow instance runs and reaches an Agent Interaction node, DAVE resolves the agent's configured provider, retrieves the encrypted key, and calls the provider's API directly. The call goes from DAVE's infrastructure to the provider. There is no intermediate proxy charging a percentage. The provider bills your account at their published rate for the tokens consumed. DAVE does not see that bill and does not add to it.

This means the economic relationship is: you pay voolama for the DAVE platform (a flat subscription), and you pay the AI provider for the model calls (at their published rate, directly). The two costs are independent. Scaling your AI workflow volume increases your provider bill and does not increase your DAVE bill beyond the seat tier you are on.

For teams running multiple providers side by side, each provider is a separate record in DAVE with its own key and model catalog. You can route different agents through different providers within the same workflow, and each provider bills your account directly at their own rate. DAVE does not aggregate or mark up across providers.

The practical objection to BYO-keys is key management: if you hold the keys, you are responsible for rotating them, securing them, and handling the consequences if one is compromised. DAVE is designed to make this manageable.

API keys are stored in a per-tenant encrypted vault using AES-256-GCM authenticated encryption. They are masked by default in the Admin UI (click the eye icon to reveal temporarily) and are never returned through the API once stored. Keys are transmitted only over HTTPS and are never logged in plain text.

Rotation is centralized. Because all agents that use a given provider reference the same Provider record, rotating a key means editing one record at Admin, then API Management, then Providers, entering the new key, and saving. Every agent that references that provider picks up the new key on its next invocation. You do not need to update individual agents, workflows, or workflow versions.

The Providers page is restricted to the Admin role. Users with Create, Curate, Use, or Reporting roles cannot view, create, edit, or delete providers. This means key management is a controlled operation: only the users your administrator designates as Admins can touch provider credentials.

After rotating a key, use the Fetch Models button on the provider card to verify that the new key authenticates successfully with the provider before your next workflow run. A successful model fetch confirms the key is valid and the provider is reachable.

The cost difference between a markup-based platform and a BYO-keys platform is not theoretical. It is calculable from your actual usage data. If you have access to your current AI platform's billing breakdown, the comparison is straightforward.

Start with your total AI spend on your current platform for a representative month. Separate the model cost (what the provider would charge at their published rate) from the platform's markup (the difference between what you paid and what the provider charges). That markup figure is the recurring cost you eliminate by moving to a BYO-keys platform.

Then add the DAVE platform cost (a flat subscription based on your seat count) and compare the two totals. For teams with low AI usage volume, the difference may be small. For teams with significant workflow automation, the markup line is often the larger number.

Two additional factors are worth including in the comparison. First, model flexibility: on a markup-based platform, the provider selection is often limited to what the platform has negotiated or chosen to support. On DAVE, you connect to any of the seven supported provider types, including OpenAI-compatible endpoints and local models via Ollama, which means you can optimize for cost and capability independently. Second, cost transparency: with BYO-keys, your AI spend is visible directly in your provider's billing dashboard, broken down by model, by key, and by time period. You can see exactly what each model costs and optimize accordingly.

The goal is not to minimize AI spend. The goal is to ensure that AI spend buys you model capability, not platform overhead. Those are different things, and the difference compounds as your usage grows.