Intelligence / Advisor

From messy usage to a short queue of fixes.

Advisor reads your usage, finds the waste, and explains each finding with evidence. Approved Autopilot execution closes the loop after review - never silent automation.

Start free →See how it works →
Free to start · no credit card · nothing changes without your approval
What Advisor catches

The waste hiding in your usage.

Across Claude Code, Codex, Cursor, Pi and your cloud bills, Advisor looks at the whole picture - not one tool at a time. It watches the places waste hides and keeps the evidence attached to every recommendation.

Usage detectors
  • ✓quota_window_reset Quota and reset windows that change the cost of where work should run next.
  • ✓model_cost_cache_regression A pricier model crept in, or cache reuse dropped and tokens are being re-billed.
  • ✓Long-context session cost. Sessions dragging huge context windows turn-over-turn at premium rates.
Autopilot decisions
  • ✓AI-spawned-session cost. Sub-agents and background runs multiplying spend under one parent task.
  • ✓Plan fit. Demote, promote, or switch only when the account evidence supports it.
  • ✓Approved execution and rollback. Apply reviewed fixes, record what changed, and track rollback state.
A
Advisor queue14 days analyzed · 6 accounts · 4 tools
+$1,040/moif you approve all
↺
Quota window reset changed the best place to run work

detector Codex API spend is rising while a subscription window has room.

+$420/mo
ReviewSkip
⚡
Cache reuse regressed on the API key

detector Prompt-cache hit rate fell 71% → 22% after a prompt change.

+$280/mo
ReviewSkip
◧
Plan fit detector

autopilot Compare seats and tiers against how each account is actually used.

+$140/mo
Track
⌥
Long-context and spawned-session detectors

detector Attribute context waste and agent-spawned work back to the parent task.

+$200/mo
Track
Evidence, not vibes

Every finding shows its work.

No black-box "trust me." Each recommendation comes with the proof, so you can check it before you act on it.

ottto · advisor / recommendation #R-412
Cache reuse regressed on the API key
Bottom line+$280/mo by restoring the cached prefix
What changedPrompt-cache hit rate fell from 71% to 22% on Jun 3.
Confidenceexact repriced from cloud actuals
CaveatsAssumes current token volume holds for 30 days.
cache reuse 22% token volume 1.9B / mo model mix opus · sonnet pricing health fresh cloud actuals reconciled
✓ A governed live agent used read-only data tools and a sanitized calculation tool, verified each claim deterministically before showing it, and stayed unresolved rather than guessing where evidence ran out.
Illustrative recommendation detail.
Advisor → Autopilot

It finds it, explains it, estimates it, and asks.

Advisor never acts on its own. You review the findings you trust, request help on the ones you want a human on, or skip the rest. Approved Autopilot execution records the change, the audit trail, and the rollback state.

Analyze every dayfull usage + cloud
→
Flag anomaliesregressions & spikes
→
Recommend with the $bottom line + proof
→
You reviewexplicit approval
→
Track real savingsverified after
Tracked savings

Closed loop, not a promise.

Expected vs verified vs disproven. The Advisor loop is built around measuring what happened after a recommendation so you know what actually landed - not just what was projected on the day you reviewed it.

  • ✓Expected. The estimate at approval time, with its confidence label.
  • ✓Verified. Re-measured against real bills and telemetry after the change.
  • ✓Disproven. Honestly marked when the savings didn't materialize - no quiet rounding up.
$1,210verified saved · last 90 days
Verified$1,210
Expected, in progress$510
Disproven$230

Turn your usage into a short queue of fixes.

Advisor brings you findings with the dollars and the proof. Review what you trust, approve the fixes, and keep the execution trail tied to measured savings.

Start free →See how it works →
Free to start · no credit card · nothing changes without your approval