The two costs, kept separate
Almost every AI project has a one-time build cost and an ongoing usage cost, and confusing them leads to bad budgeting. The build is the engineering to design, connect, test, and ship the feature. The usage is what you pay the model provider each time the feature runs.
We quote and track them separately so you can see both clearly. The build is a project; the usage is a utility bill that scales with how much people use the feature.
- Build cost: scoping, integration, testing, and deployment — paid once
- Usage cost: per-request model fees (tokens) that scale with volume
- Maintenance: optional ongoing support as your needs change
- Usage is billed by the provider (OpenAI, Anthropic, etc.), not marked up in the dark
- We set usage caps so the ongoing number stays predictable
What drives the price
Build cost is driven by scope: how many workflows, how much custom logic, how deeply it connects to your existing systems, and how strict the accuracy and security requirements are. A single, well-defined task is far cheaper than a broad platform.
Usage cost is driven by how often the feature runs, how much text goes in and out, and which model you choose. A bigger model or long-document context costs more per call — sometimes worth it, sometimes not.
- Number and complexity of workflows in scope
- Depth of integration with your current stack and data
- Accuracy, security, and compliance requirements
- Volume of requests and size of each prompt/response
- Model choice — larger and long-context models cost more per call
- Whether you need ongoing maintenance and iteration
More on ai integrations
Frequently asked questions
Can you just tell me a price?
Not honestly without knowing the scope — a one-page assistant and a multi-system agent are very different projects. What we can do fast is scope a specific starting slice in a free consultation and give you a clear, fixed number for that.
Why does the model usage cost change month to month?
Because it scales with use — more requests and larger prompts mean more tokens billed by the provider. We set per-user and per-day caps so it stays in a predictable range and never surprises you.
How do we keep costs down?
Right-size the model to the task, trim prompts, cache stable content, and start with one high-value workflow instead of building everything at once. We design cost control in from the start rather than bolting it on later.