Last month your Claude or OpenAI invoice doubled and nobody can tell you why. Your Max plan hits rate limits mid-sprint. Your team runs Opus for everything because switching feels risky. Adapterly fixes all three — and the agent uses the data to optimize its own behavior, guided by battle-tested cost patterns baked into the skill.
01 · The runaway session
$400 in one session? Find it in 10 seconds.
Context bloat, silent background watchers, misrouted models — flagged automatically with the session id, the model, the day it happened.
02 · The rate-limit wall
72% of your Max quota. 1h 18m to reset.
Live headroom dial per model tier for Pro / Max / Team flat plans. See which workflow is about to cut you off, before it does.
03 · The agent that optimizes itself
Not just visibility — action.
Your agent reads its own cost data and applies battle-tested patterns from the skill — model routing, context splitting, cache usage — and proposes concrete cuts: “Switch routine edits to Sonnet. Save ~$187/mo.” Always with your approval, benchmarked first.
The numbers that answer "why?"
Rate-limit headroom · last 5h · Max 5×
Frontier (Opus)12 / ~50
Mid (Sonnet)163 / ~225
Small (Haiku)22 / ~500
Oldest call 3h 42m ago · first slot frees in 1h 18m. floor estimate
Top sessions · last 7d
8d832a1…claude-opus-5-5 · context bloat$227.63
e85930f…claude-sonnet-5-5 · interactive$113.26
d056564…opus-5 · silent watcher$36.52
7bd2a84…sonnet-4-6 · mixed$18.09
Click a session for per-operation + per-day trend.
The agent's playbook (built into the skill):
pre-flight cost check · tier-aware model routing · context bloat / silent-watcher detection · cache-opportunity flagging · approval-gated benchmarking before any switch.
“I don't know which session is burning my quota — and claude.ai doesn't tell me.”
The engineering lead with a $2k+/mo invoice
“I need per-session, per-dev attribution and routing advice I can defend to finance. In the EU.”
Free during beta. One week to proof.
Point your OTel pipe at us. By Friday you'll know which session, workflow, and model is draining your budget — and have a plan to cut it. EU-hosted in Hetzner Falkenstein. Metadata only, never prompt bodies. adapterly.ai/ingest · hello@adapterly.ai
Not contractual. Pricing & plan quotas are approximations; verify before quoting. v1.3 · 2026-10-27