Agent Analyzer
AI Agent Observability

Understand every
AI agent run.

Cost. Tokens. Latency.
Problems.

Analyze an AI agent run and see exactly where your tokens, money, and latency went — without digging through logs.

Built for developers building with LLMs
OpenAIAnthropicGeminiLangGraph

Your agent works. But do you know what it's costing you?

One user request can trigger multiple model calls, tool calls, retries, and increasingly large contexts. The final answer may look simple while the execution underneath is anything but.

7LLM calls
4tool calls
37,031tokens
8.73sduration
$0.284cost

From chaos to clarity

Without analysis

server.log
request started
llm request
tool call
llm request
retry
llm request
tool call
llm request
...
request completed
8.73s$0.284
¯\_(ツ)_/¯

With Agent Analyzer

run-analysis.json
COST$0.284
BIGGEST COSTStep 06 — 48%
CONTEXT GROWTH7.2×
REPEATED TOOLsearch_orders ×4
POTENTIAL ISSUELarge repeated context

What we analyze

Everything that matters
in an agent run.

01

Cost

Track total cost and cost per step. Find which step is burning your budget.

02

Tokens

See exactly where input and output tokens are going across every LLM call.

03

Latency

Find the slowest steps in an agent run. Pinpoint exactly where time is lost.

04

Tool Calls

See which tools were called, how often, and whether any are being called unnecessarily.

05

Context Growth

Detect rapidly growing context that inflates token costs on every subsequent call.

06

Problems

Automatically surface potential inefficiencies — before they become expensive habits.

Not just numbers. Explanations.

Raw traces tell you what happened. Agent Analyzer helps you understand why it happened.

run-analysis — customer-support-agent

Run Analysis

customer-support-agent

3 Potential Issues
01HIGH CONTEXT GROWTH

Context increased 7.2× during this run.

Potential impact:Higher input-token cost and latency.
Recommendation:Summarize previous context before subsequent calls.
02EXPENSIVE FINAL STEP

Step 06 represents 48% of total run cost.

Potential impact:$0.137 of $0.284 total cost.
03REPEATED TOOL CALL

search_orders was called 4 times.

Recommendation:Review whether results can be cached.

How it works

Three steps from run to insight.

01

Bring a trace

Upload a supported agent trace or paste a run. We accept common formats from popular frameworks.

02

We analyze it

We calculate cost, tokens, latency, calls, context growth, and potential issues — automatically.

03

Fix what matters

See where your agent is spending time and money and what to investigate next.

Developer-first

Get answers without
the overhead.

No complicated setup. Bring a single agent run, analyze it in seconds, and see exactly where your tokens, time, and money went.

NO SETUP REQUIRED

Go from trace to actionable insights without configuring a monitoring stack.

START WITHOUT AN ACCOUNT

Analyze your first run immediately. No signup wall, no commitment.

WORKS WITH YOUR STACK

Designed to work with the tools and models developers already use.

Find out what your agent
is really doing.

Stop guessing where the tokens, time, and money went.

Free to try