Craig Credits and Pricing

Craig uses a pay-as-you-go credit model. There are no plans to purchase, no subscriptions to manage, and no minimum spend. You only pay for what you use.

What is a credit?

A credit is the unit used to measure Craig's usage. Each time a Prompt is run or an Agent checks a condition, credits are consumed based on the complexity of the task and the model assigned. Credits are billed in increments of one tenth of a credit — so you pay only for what Craig actually uses, not a full credit rounded up.

For current credit pricing, see connectrocket.com/pricing.

How credits are consumed

The number of credits consumed by a Prompt or Agent depends on two things: the complexity of the task and the model Craig uses to complete it.

Craig assigns one of two models to each Prompt and Agent after an initial discovery phase:

• Fast — used for straightforward retrieval tasks. Lower credit consumption.
• Thinking — used for tasks that require conditional logic or reasoning. Approximately three times the credit consumption of Fast for the same task.

The model assigned to each Prompt and Agent is displayed in the UI alongside an estimated credit cost per execution. This gives you a clear picture of what each run will cost before it happens.

For more detail on how model assignment works, see Fast and Thinking: How Craig Chooses the Right Model.

How prompt and agent design affects your costs

Because Thinking uses approximately three times the credits of Fast, how you write your Prompts and Agents has a direct impact on your costs.

A Prompt or Agent with complex conditional logic — if/then rules, comparative evaluations, multi-step reasoning — is more likely to be assigned the Thinking model. A simpler, more direct Prompt or Agent that retrieves and returns values without conditional logic is more likely to be assigned Fast.

This means that a poorly drafted Prompt or Agent can cost significantly more to run than a well-drafted one performing the same task. Revisiting and simplifying a Prompt or Agent after initial configuration can result in Craig reassigning it to the Fast model — reducing your credit consumption going forward.

Example:

A Prompt asking Craig to fetch the current river level and return it plainly is likely to be assigned Fast. The same Prompt rewritten to include conditional formatting rules — show this section only if the level exceeds a threshold, apply different formatting depending on the trend — is more likely to be assigned Thinking, at approximately three times the cost per run.

If you review the model assigned to a Prompt or Agent and find it's running on Thinking when you expected Fast, revisit the instructions for conditional language, conflicting rules, or unnecessary complexity. Simplifying the Prompt or Agent and running Save & Verify again will trigger a new discovery phase — and may result in Fast being assigned going forward.

See Common AI Mistakes to Avoid When Writing Craig Prompts and Writing Good Conditions for Craig Agents for guidance on writing efficiently.

Agents and check frequency

For Agents, credit consumption is determined by both the model assigned and how often the Agent checks its conditions. An Agent checking every five minutes consumes credits far more frequently than one checking every hour.

When configuring check intervals, match the frequency to how quickly the underlying data actually changes. A river gauge that rises over several hours doesn't need the same check frequency as a fast-moving storm front spawning tornados. Checking more often than the data updates wastes credits without improving the Agent's responsiveness.

See Configuring Agents for guidance on setting check intervals.

Viewing your credit usage

Credit usage appears as a line item on your Connect Rocket invoice. There is no separate credits dashboard — usage is visible at billing time, alongside your other account charges.

The estimated credit cost for each individual Prompt and Agent is displayed in the UI at the time of configuration, giving you a reference point before you commit to running at scale or at high frequency.

Using caching to reduce Prompt costs

Prompts support a Cache Duration setting that can significantly reduce credit consumption when the same Prompt is run by multiple users within a short period.

When a cache duration is set, Craig runs the Prompt once and stores the result for the defined window — for example, 15 minutes. If any other user runs the same Prompt within that window, they receive the cached result instantly at no additional credit cost. The Prompt can be called once and served to many users without consuming any further credits until the cache expires and a fresh execution is required.

This is particularly valuable during incidents or activations where several team members may run the same situational awareness Prompt within minutes of each other. One credit is consumed for the initial execution — every subsequent run within the cache window is free.

Cache Duration is configured within each Prompt individually. See Configuring Prompts for guidance on setting an appropriate duration for your data type.

Caching is best suited to data that changes slowly relative to how often your Prompt is run — tide charts, avalanche forecasts, or daily weather summaries. Do not cache rapidly changing data such as severe weather alerts or tsunami warnings, where a stale result could have operational consequences.

Agents do not have a caching feature — each condition check is a live execution.

Summary

^{*Subject to change}

The most effective way to manage Craig costs is to write clear, direct Prompts and Agents, set check intervals that match the rate of change of the data you're monitoring, and review the assigned model after any significant edits.