Overview: the number your bill cannot give you
The first page joins two streams your stack keeps separate: what your AI features cost and what your users pay. Total cost, total revenue, and margin, plus the trend. This join is the entire premise: a bill going up is not a problem or a success by itself, which is the diagnosis problem covered in the three warning signs. Every number is computed server side from tokens at current prices, per the methodology.
Users: sorted by margin, worst first
The page that finds your quietly unprofitable customers. Each row is one user: their plan, their revenue, their LLM cost this period, their margin. Sort ascending and the users costing you more than they pay are at the top, the exact one minute walkthrough in finding your first unprofitable user. On the demo the rows are fictional ids like u_412; on your project they are your real users the day you integrate.
Features: which product surface burns the money
Cost broken down by the feature label you attach to each call: chat versus summarize versus your agent, with per feature model mix. This answers the pricing question, which features can live inside your current plans and which need their own limit, the analysis behind whether an unlimited AI plan survives.
Loops and alerts: the runaway record
- Loops shows token velocity anomalies, users who burned past the threshold (default 50,000 tokens in 5 minutes), the runaway agent signature explained in the detection flow.
- Alerts is the delivery history: what fired, when, to Slack or email, so 2am incidents have a record you can audit the next morning.
Pricing, recommendations, settings: the levers
The pricing page is where caps live, per plan monthly budgets with block or downgrade behavior, the decision covered in block or downgrade. Recommendations analyzes your actual usage and suggests cheaper model swaps and pricing adjustments. Settings holds alert thresholds, the Slack webhook, and key rotation. Everything here is no code: set a cap in the dashboard and the SDK enforces it before the next call.
FAQ
What should an LLM cost dashboard show?
Not just total spend, that is what your provider bill already shows. The views that change decisions are per user cost joined to per user revenue (margin), per feature cost so you know which product surface burns money, loop and anomaly history so runaways are visible, and enforcement state, which users are capped and what happened when they hit the cap. Weckr’s dashboard is built as exactly those views.
Can I see the Weckr dashboard without signing up?
Yes. useweckr.com/demo is the full dashboard running on seeded fictional data, every page, no signup and no card: overview, users, features, loops, alerts, pricing caps, recommendations, and settings. The users and numbers there (ids like u_412, projects like Acme) are generated, disclosed as such, but the views and math are the real product.
What is the difference between a cost dashboard and margin tracking?
A cost dashboard alone tells you what you spent; margin tracking joins that spend to what each user pays you. The second is where decisions live: a $400 monthly OpenAI bill is neither good nor bad until you know whether it comes from profitable Pro users or three free tier users running loops. Weckr stores each user’s plan revenue with their usage so every page can show margin, not just cost.
How does data get into the dashboard?
From a two line SDK wrap around your existing OpenAI, Anthropic, Gemini, or Kimi client. Each call logs tokens, model, user id, feature, and plan; cost is recomputed server side at current prices; the dashboard reads those rows live. Integration is about ten minutes and there is a walkthrough article covering the exact steps.
Does the dashboard update in real time?
Rows appear as calls are logged, so the dashboard reflects usage as it happens rather than on a billing cycle delay. That is what makes loop detection useful: a runaway agent shows up in the loops view and fires an alert within minutes, not on next month’s invoice.
Keep reading
Take the tour, then fill it with your data
The whole dashboard is open at useweckr.com/demo right now, seeded data, every page. When the views earn it, the ten minute integration replaces the fictional users with yours, free for 50,000 requests a month.