Your AI stack, understood.
See where the work goes. Turn one insight into a measured improvement.
Reading your real usage…
Native activity is not a subscription bill. OpenRouter spend is reported separately.
Finding a useful place to start…
The shape of your work
Select a day to explore the full usage view.
One stack. Different signals.
Native tools report local activity. OpenRouter reports this API key’s spend. These are not added together.
High-volume projects · where to investigate
Start with high-volume project groups. Each suggestion is a hypothesis to test, not proven waste.
Across your tools
NAME MATCHES · UNCONFIRMEDSpot project names appearing in both tools. Inspect each source before treating them as the same work.
Project decision workspace
REAL METADATASELECTED PROJECT
Connecting your stack…
Observed activity → a specific hypothesis → a measured test.
01 / Model mix
OBSERVED TOKEN SHAREModel attribution appears after collection.
Model a smaller workloadYOUR ASSUMPTIONS · EXPAND
What if this project needed less new context or shorter answers? Model the possibility before changing anything.
Cached input stays unchanged. Reasoning is already part of output. Subscription fees may stay the same; dollar savings are unknown.
How to prove an improvement
From an idea to an improvement
PROVE THE OUTCOMEThe brief gives an AI agent the observed evidence, a proposed experiment and quality checks. No model has been changed or tested by this dashboard.
Compare the work you plan to run.
Enter a text workload, select models, then inspect assumptions and missing fees.
Select up to 4 models to compare.
See what changed. Find where to look.
Compare complete UTC days across your local AI tools. Today stays separate.
Open this view to read your usage history.
Daily observed activity
COMPLETE UTC DAYSChoose a day for its exact observed totals.
Empty days mean no accepted records in this snapshot. Partial collection can hide activity; historical backfill can change the shape of this chart.
View daily figures
| UTC day | Updates | Tokens |
|---|
Projects to investigate
OBSERVED VOLUMEDirectory attribution is unconfirmed. Volume ranks investigation opportunities; it does not measure work quality.
Models doing the work
SOURCE + MODELObserved model labels do not establish price, capability or the best alternative for a task.
OpenRouter spending check
CONFIGURED KEY / USDCompare this key’s reported month-to-date spending with a limit you choose.
No reference limit set.
Local to this view; not saved. This does not set a provider limit or stop spending. OpenRouter key spend is separate from subscription activity, account credits and the date filters above.
How much can we conclude?
CHECKING COVERAGEYour source coverage will appear after analysis.
Your improvement plan.
Small, controlled tests. Original baselines. Outcomes you can inspect.
Did the change actually help?
Reopen a review, capture a task from your real tools, then mark whether you accepted the result. Usage and available request charges are recorded automatically.
Supports review-case JSON and exported review tests. Files are read locally; no account connection or reanalysis.
Start a live test or reopen a review.
Create an empty review to capture real tasks automatically, or open a saved review to retain its original context. No sample measurements are filled in.
Original review
UNVERIFIED SNAPSHOT · NOT RECALCULATEDInspect recorded context and requirements
Manual measurements · optional alternative
Same work. One deliberate change.
Count each original task once. Include billed retries and variable fees in spend. Fixed subscription commitments are separate. Leave unknown measurements blank.
Project hypotheses
EXISTING IMPROVEMENT PLANWhat counts as an improvement?
Use the same representative tasks and acceptance criteria. Record original tasks once; include retries in the total variable cost. Keep quality, latency, fixed subscriptions and variable charges separate. A lower cost alone does not establish equivalent results.
Know what is connected. Know what is missing.
Collection health, provider permissions and privacy in one place.
Claude Code
CONNECTINGDiscovering local usage metadata…
Tokens do not establish actual charges.
Codex
CONNECTINGDiscovering local usage metadata…
Tokens do not establish actual charges.
OpenRouter · USD
CONNECTINGChecking reported key counters…
Key scope and account credit scope stay separate.
Scope · privacy
METADATA ONLYNo estimated subscription charges or fabricated savings. OpenRouter counters are reported separately.
Local session metadata is read on this machine. No raw conversations are displayed or exported.