Usage
The Usage page under Admin → Organisation → PebbleObserve is the org-wide spend and activity dashboard. It’s the page you check first thing every morning to know whether anything blew up overnight.

Find it at Admin → Organisation → PebbleObserve → Usage.
What you see at the top
- Usage View dropdown — switch between
Global Usage(whole org) and other scopes if available - Select Time Range —
Last 7 Days,Last 30 Days, custom range - Export Data button (top-right) — download the current view as CSV
- Project Spend Apr 8 – Apr 15, 2026 (Organisation) — total spend in the selected window. Headline metric.
- Total Spend — dollar amount (e.g.
$13.3856) - Max Budget —
No limitif no budget has been set, otherwise the configured cap
The five perspective tabs
Below the headline numbers, five tabs let you slice the same time window five different ways:
Cost
The default view. Shows total spend with a Daily Spend bar chart underneath.
This is your “is anything weird happening with money” view. A spike on a particular day usually means either:
- A user ran an expensive multi-step agent
- A flow has a runaway loop calling an expensive model
- Someone published a flow with a Large Language Model node configured for the most expensive option
Use the daily chart to spot anomalies, then jump to Logs for that day to investigate.
Model Activity
Cost broken down by model. Tells you which models are doing the heavy lifting (typically Claude Opus and Sonnet, with Haiku doing volume but lower cost).
Use this to:
- See whether your routing is doing what you expect (lots of Auto traffic but routed mostly to Haiku = good cost optimisation)
- Identify models that are barely used (candidates for removing from your catalogue)
- Spot expensive models being used for cheap tasks (a sign your routing strategy needs tuning)
Key Activity
Activity broken down by API key. Each PebbleRouter API key has its own row, so you can see which application or integration is using how much.
Use this to:
- Identify a runaway client driving up cost
- Check whether a new application’s usage matches expectations
- Track usage of a specific team’s API key
MCP Server Activity
Activity broken down by MCP server (Microsoft 365, Jira, custom MCPs). Tells you which integrations are most-used.
Use this to:
- Justify keeping (or removing) a paid integration
- See which integrations are getting traction with users
- Spot integrations that are misconfigured (zero usage despite being enabled)
Endpoint Activity
Activity broken down by API endpoint (chat completion, embedding, image, etc.). Tells you what kind of work PebbleAI is doing.
Use this to:
- See whether embedding usage is dominating cost (a signal you might need a cheaper embedding model)
- Track image-generation usage if your org has it enabled
- Verify the balance of cost between chat and other API surfaces
Common workflows
”What did we spend last week?”
- Set time range to
Last 7 Days - Stay on the Cost tab
- Read the Total Spend number
- Optionally export as CSV for the finance team
”Why did spend spike on Tuesday?”
- Set time range to last 30 days
- On the Cost tab, find Tuesday on the bar chart and note the value
- Switch to Model Activity to see which model drove it
- Switch to Key Activity to see which client did it
- Jump to Logs, filter by Tuesday + the offending model, find the actual conversation
”Is my routing actually saving money?”
- Time range:
Last 30 Days - Model Activity tab
- Look at the proportion of total spend going to your Fast model (should be high) vs Smart model (should be low)
- If too much is going to Smart, your routing strategy needs tuning — see PebbleRouter
”Which integrations are actually being used?”
- Time range:
Last 30 Days - MCP Server Activity tab
- Sort by usage; integrations near the bottom are candidates for removal in Capabilities
Step-by-step: setting a budget
Budgets aren’t set on this page directly — they’re configured via the API or in your platform admin settings. If your install supports budgets, the Max Budget field in the header will show the cap; if it shows No limit, no budget is configured.
Talk to your platform admin to set or change a budget.
Exporting
The Export Data button (top-right) downloads the current view as CSV — useful for:
- Finance team reports
- Sharing usage data with stakeholders
- Long-term tracking outside PebbleAI
The export includes only the currently visible data (the time range you’ve selected, the perspective tab you’re on).
How this relates to user-facing usage
Each user can see their own usage at Profile → My Usage (in the avatar fly-out). That view is scoped to just them — the admin Usage page on this admin section is the org-wide view.
If a user complains that PebbleChat is “expensive”, check their My Usage view first — they can usually see for themselves which models they’re picking, and a quick conversation about model choice often solves it.
Tips
- Make this a daily habit. Five seconds checking the Cost tab catches most weirdness before it becomes a problem.
- Set up a weekly export for the finance team if your org needs invoicing visibility.
- Watch the model mix. A healthy mix is heavy on Fast model, light on Smart model. Inverted means routing is misconfigured.
- Use Key Activity for capacity planning. If a single client API key is approaching limits, it’s time to talk to the team using it.
Related
- Logs — drill from a usage anomaly into individual requests
- Evaluations — measure quality, not just cost
- PebbleRouter — tune the routing that drives the model mix
- PebbleObserve (user-facing) — the deeper observability platform behind these dashboards