Understand what your
agents are actually doing
Per-agent cost rollups, retry-storm alerts, unused-tool flags, and week-over-week spend deltas, all surfaced automatically from your existing proxy traffic. No instrumentation, no SDKs.
Per-agent cost rollups
Every agent is fingerprinted by its system prompt. RelayPlane groups spend by agent identity and shows you exactly which agent burned what, this week and last.
Retry-storm detection
When an agent retries the same call more than 3× in a session, Insights flags it as a retry storm. You see it the moment it happens, not after the bill arrives.
Unused-tool flagging
Tools declared in a system prompt but never called inflate token counts for every request. Insights surfaces which agents are carrying dead weight so you can trim it.
Week-over-week delta
See whether each agent is getting cheaper or more expensive. WoW delta is only shown when an agent has enough traffic to be statistically meaningful (5+ requests).
What you'll see
Three live views computed from your proxy traffic, no configuration required once RelayPlane is running.
Every agent ranked by spend. WoW delta shows whether it's trending up or down.
Retry storms, unused tools, and cost improvements flagged in real time.
Daily spend bars for this week vs last. Total delta at a glance.
How it works
Your agents call RelayPlane
Point any agent at localhost:4100. The proxy intercepts every request, no changes to your agent code.
Insights runs automatically
System prompts are fingerprinted to identify agents. Cost, retry patterns, and tool usage are computed from the raw request log, no instrumentation needed.
Open the dashboard
Your Insights dashboard updates live. Week-over-week deltas reset every Monday. Pattern alerts surface as they happen.
Built for agent developers who hate surprises
If you run more than one agent, a code reviewer, a PR summarizer, a test generator, you've probably been flying blind on which one is eating your budget. The Insights dashboard gives you the view you've been missing: not just total spend, but per-agent attribution, trend direction, and the patterns that signal something is about to go wrong.
- See which agent is the most expensive at a glance
- Catch a retry storm before it multiplies your bill
- Trim unused tools to reduce overhead on every request
- Know immediately if a code change made your agents cheaper or more expensive