Real-Time Cost Tracking
SuperBuilder gives you complete visibility into what every AI task costs. Per message, per thread, per project — the cost is always visible, updated in real time, and never a surprise.
The Problem with AI Cost Blindness
Most AI tools show you a monthly invoice. By then, it's too late to do anything about it. You don't know which tasks were expensive, which models you should have used instead, or where your budget went.
SuperBuilder tracks cost at the granularity of individual messages. You see the cost of each response as it streams in, the running total for the current thread, and the aggregate across your workspace. You can make smarter decisions in the moment — not after the fact.
What Gets Tracked
Per-message cost — every response from the AI includes its token count and dollar cost. You see this live as the response generates.
Thread total — the running cost of the current conversation, updated with each exchange.
Project total — aggregate spending across all threads in a project, so you can see which projects are consuming the most budget.
Model breakdown — costs are tracked per model, so you can see what you're spending on Opus vs Sonnet vs Haiku and make informed choices.
Seeing Cost in the UI
Cost indicators appear throughout SuperBuilder:
- In the thread view — a small cost badge appears next to each AI response showing its cost
- In the thread list — each thread shows its total spend
- In the project sidebar — project-level totals give you a quick overview
- In the cost dashboard — a dedicated view shows spending trends, model breakdowns, and per-day totals
The numbers are never hidden behind a settings menu. They're front and center because they should inform every decision you make.
Budget Alerts
You can set budget thresholds at the project or workspace level. When spending approaches your threshold, SuperBuilder notifies you before you exceed it — not after.
Alerts fire via:
- In-app notification
- Push notification on your phone
- Any notification channel you've configured
This is especially useful when you have agents running unattended overnight or through the scheduler. You set a budget, and you're guaranteed not to wake up to a surprise bill.
Making Cost-Conscious Decisions
Cost visibility changes how you work with AI:
Model selection — when you can see that a Haiku response costs 1/20th of an Opus response, you naturally reach for the right tool. Exploration and quick questions go to Haiku. Heavy refactors go to Opus.
Task scope — you learn what scope of task produces what range of costs. A "refactor this file" task might run $0.08-0.40 depending on complexity. Knowing this helps you batch work intelligently.
Agent termination — if an agent is going in circles and you can see the cost climbing, you know to stop it and redirect rather than let it keep burning tokens.
Thread Cost Reports
At the end of any thread, you can export a cost breakdown:
- Total tokens input and output
- Cost per model used
- Average cost per message
- Comparison against previous similar threads
This is useful for teams that need to charge back AI costs to specific projects, or for understanding how your usage patterns are changing over time.
Frequently Asked Questions
Does SuperBuilder add a markup to AI costs?
No. SuperBuilder passes through costs at whatever rate your AI provider charges. The cost you see in the UI matches what you're charged by Anthropic, OpenRouter, or your configured provider.
Can I set different budgets for different projects?
Yes. Budget thresholds are configured per project and per workspace independently.
What happens when I hit my budget limit?
By default, SuperBuilder sends a warning alert but continues running. You can configure it to pause new agent tasks when the limit is reached — this is the safer option for unattended automation.
Can I see historical spending over time?
Yes. The cost dashboard shows spending by day, by week, and by month. You can filter by project, model, or thread type.
Does tracking affect performance?
No. Cost tracking is computed from token counts that are already reported by the AI provider API. There's no additional computation or latency.
