Bring your own keys: how BYOK changes AI cost control

Most AI platforms bundle model usage into their subscription price. You pay one fee and hope the margin is reasonable. The problem is you never see the actual AI cost, so you cannot optimize it. You are paying a markup on every API call, and the platform decides which model runs each task.

ProDexter takes a different approach. Instead of marking up AI usage, we separate the two costs entirely. The platform subscription covers automation, connectors, and orchestration. AI usage goes straight to your own provider account. This is BYOK: bring your own keys.

The result is full visibility into both sides of the bill. You control the platform cost by choosing a plan that fits your team size. You control the AI cost by choosing models, managing context, and watching usage in your own dashboard.

What BYOK means

BYOK stands for bring your own keys. When you connect ProDexter to an AI provider, you enter your own API key. Every time an agent in your workflow calls a language model, the request goes through your provider account, not ours.

This means ProDexter never touches your AI spend. We do not proxy requests, add fees per token, or take a percentage. The provider bills you directly at their published rates. You see every call, every token count, and every dollar in your own provider dashboard.

You can use keys from multiple providers at the same time. If you have accounts with different AI providers, ProDexter can route different steps to different keys. There is no lock-in to a single vendor on the AI side.

Why it matters for cost visibility

When AI usage is bundled into a platform fee, you cannot tell how much of your bill is automation and how much is model inference. If costs rise, you do not know whether you need fewer workflows or cheaper models. The two expenses are tangled together.

BYOK creates a clean separation. Your ProDexter invoice covers the platform: workflow orchestration, native connectors, scheduling, team seats, and support. Your AI provider invoice covers model usage: tokens consumed across all your workflows. Two line items, two dashboards, two levers you can pull independently.

This separation makes budgeting straightforward. Finance teams can allocate platform cost to the operations budget and AI usage to a separate line. Department heads can track their own AI consumption without asking the platform vendor for a breakdown.

Choosing the right model for each step

Not every step in a workflow needs the most capable model. A task that extracts a date from an email does not require the same model as one that drafts a legal summary. Running every step on the largest model wastes money on tasks that a smaller, faster model handles just as well.

ProDexter routes each task to the right model, agent, or employee. Simple classification, extraction, and formatting steps can run on smaller models. Complex reasoning, generation, and analysis steps route to more capable ones. You set the rules, or let the optimizer suggest them based on task complexity.

Some steps do not need AI at all. ProDexter routes work to the right handler for each task: an AI model, a rule-based agent, or even a human review step. The platform decides the routing. You decide the policy. The bill reflects only what was actually needed.

What still drives the bill

Even with the right model for each step, one factor dominates AI cost: context size. Language models charge by the token, and both input and output tokens count. The more context you send with each request, the more tokens you consume, and the higher the cost per call.

Context adds up fast. If a workflow step receives an entire document when it only needs one paragraph, you are paying for tokens the model reads but does not use. Multiply that by hundreds of workflow runs per day and the waste compounds.

Large context windows also increase latency. More tokens in means more time processing, which slows down your workflows. Keeping context tight is not just a cost decision. It is a performance decision.

How context selection keeps costs down

ProDexter uses context compression to send only the relevant information to each step. Instead of passing an entire document, conversation history, or database result to the model, the platform identifies which parts of the context are needed for the specific task and strips out the rest.

This works at every stage of a workflow. When an agent reads a long email thread, context selection extracts the latest message and key reference points rather than sending the full thread. When a step queries a database, the platform passes the relevant rows, not the entire result set. When a document needs summarizing, only the sections matching the query go to the model.

The effect is smaller, cleaner input for every API call. Fewer tokens in means lower cost per call and faster responses. Over hundreds or thousands of runs, the savings from tighter context add up without any change to output quality.

The pricing model

ProDexter’s plans cover the platform only. AI usage is always separate, billed by your provider.

The Free Sandbox at $0 per month lets you build and test workflows before committing. Starter at $29 per month covers individuals and small teams getting started with automation. Team at $99 per month adds collaboration features and higher workflow limits. Growth at $299 per month is built for scaling teams with heavier usage. Business at $799 per month includes advanced controls, priority support, and higher concurrency. Enterprise starts at $1,999 per month and includes custom integrations, dedicated support, and SLAs.

Every plan includes BYOK. There is no premium tier required to use your own keys. The platform fee stays predictable, and your AI spend scales with actual usage under your direct control.

What this means in practice

With BYOK and context compression working together, you control both levers that determine your total automation cost. The first lever is model selection: which model runs each step, and whether a step needs AI at all. The second lever is context size: how much information each model call receives.

Pull the first lever by setting routing rules. Assign smaller models to simple tasks and reserve larger models for complex ones. Pull the second lever by letting context compression trim inputs automatically, or by configuring context rules for specific workflow steps.

The platform gives you the data to make these decisions. You can see token usage per step, cost per workflow run, and model performance across tasks. When something costs more than expected, you can trace it to a specific step and adjust.

This is what cost control looks like when the platform is not in the middle of your AI spend. No markups to negotiate. No opaque bundling. Two clear cost lines, both under your management.

Get started

See the full plan breakdown on our pricing page. If you want to estimate how BYOK and context compression affect your specific workloads, try the savings calculator with your own numbers. Start with the Free Sandbox and connect your keys in under five minutes.