That agent run cost $0.37
$0.28 of it was waste
Upload an agent log and AgentYield finds duplicate calls, bloated context, wrong model choices, and unnecessary retries — and tells you exactly what each one cost. No install. No account required.
Want continuous cost optimization? The SDK takes 5 minutes. OpenClaw users: install the skill.
Already using Helicone, Langsmith, or Langfuse? AgentYield Connectors run alongside them — they log requests, we find the waste. See how we compare →
Waste Score
74
74% of this run was avoidable spend
💰 $0.28 saved of $0.37 total
How it works
Drop your log file
JSONL, JSON, or CSV. Drag and drop. No SDK, no setup, no account.
We find the waste
AgentYield scans every event across 5 cost categories and puts a dollar figure on each one.
You get a Waste Score
A 0–100 score, a total savings estimate, and ranked fixes — ordered by what saves the most money first.
Most agent waste isn't a bug. It's a pattern.
The same five problems appear in nearly every production agent. AgentYield detects all of them, automatically.
Duplicate Tool Calls
5 searches. 3 unique result sets.
Your agent ran the same query under different phrasings and paid for all of them.
Oversized Context
89% context window used for a synthesis step
Raw results passed whole into context. Summarizing first would have cost 80% less.
Model Mismatch
GPT-4o used for sentiment classification
GPT-4o-mini handles that task identically. At 95% lower cost.
Excessive Retries
4 attempts. Valid output existed after attempt 2.
Two redundant attempts billed in full. The agent didn't check whether the answer was already there.
Redundant Reads
File summarized immediately after being written
The output was already in context. Re-reading added cost with zero new information.
One upload finds waste. The SDK prevents it.
Add two lines to your agent and AgentYield scores every run automatically. Track waste over time. Get alerts when costs spike. No log files to manage.
- Waste Score on every run — no manual uploads
- Cost trends and regression alerts over time
- 2 lines of code. Under 5 minutes to set up.
import { AgentYield } from "@agentyield/sdk";
const ay = new AgentYield({ apiKey: "ay_live_..." });
const run = ay.startRun({ agent: "research-bot" });
// your agent logic here
await run.end({ status: "success" });That's it. Every run gets a Waste Score.
Works with OpenClaw, Gemini, Claude, GPT-4o, and any agent that supports the AgentYield SDK.
Pricing
Start free. Upgrade when you need more.