Model keys and platform credits: understand the difference
Separate model-provider charges from workspace credits, and check what your workflow actually consumes before scaling it.
By Hot wasl
Published
“Bring your own key” describes how a model call is authenticated and charged. It does not mean that every part of a workflow is free. In Hot wasl, separate two questions: which account pays the model provider, and which nodes consume your workspace’s platform credits? Keeping those questions distinct makes a run much easier to budget.
Understand the two cost paths
When an AI node uses a configured platform model key, its model calls consume platform credits according to the selected model. When that node uses your own valid, attached model credential, those model calls cost zero platform credits. Your model provider may still charge your account under its own rates, limits, and terms.
Other work in the graph is separate. The Read web page node consumes platform credits even when the following summarisation node uses your key. Logic, text templating, and result output nodes do not acquire a model charge simply because an AI node is nearby. Look at each step rather than labelling the entire workflow “free”.
Attach a credential, not just a stored secret
Saving a key under Credentials does not automatically switch every AI node to it. Open the node’s settings and select the intended API credential. Review each model-calling node in a multi-step workflow. A model appearing in the catalogue is not a guarantee that your key or the configured endpoint can access it.
- Store the appropriate model-provider key in Credentials, not in a prompt or a URL.
- Select that credential in the AI node and check the chosen model.
- Run a small example with a positive workspace credit balance.
- Inspect the AI node’s logs and credit usage, then inspect any page-reading nodes separately.
Use the trace as your cost check
A successful model call using a workspace credential should have no platform-credit model charge. The trace can also identify a simulated model: without a configured model key, the app may produce a simulated response with no model credits. That is different from a paid provider request and should not be used to estimate real provider billing.
The current run-creation check requires a workspace balance above zero, including when you intend to use your own key. Attaching a key does not bypass that check. Start with one item before processing a list: nodes that fan out can execute once per item, multiplying the work and its associated costs.
- Confirm which credential each AI node actually uses.
- Count page reads as well as model calls.
- Compare recorded platform credits with the provider’s own usage records.
- Recheck costs after changing models, list sizes, or the graph.