Zentu is launching soon Get early access›

What are your AI agents actually worth?

Zentu gives you the answer — plain-English reports on what your agents do, what they cost, and the value they bring back. No engineer required.

Innovation partner of

Alibaba Cloud Innovation Partner — Where we build the future together
Inside the report

Track every dollar, every decision.

Cost, performance, and provider breakdowns — all in plain English.

Performance by agent

See how every agent's success rate evolves — week over week, side by side.

Learn more
support-triage 89.2%+5%
sales-researcher 73.4%+3%
doc-writer 66.4%+2%
onboarding-bot 50.5%−7%
billing-agent 40.5%+8%
100% 80% 60% 40% 20% 0% Apr 11 Apr 12 Apr 13 Apr 14 Apr 15 Apr 16 Apr 17

Activity by provider

Every run, decision, and tool call — broken down by AI provider in real time.

Learn more
All Runs 32.6K+12k
Tool Calls 72.3K+12k
Successful 13.2K+0.3k
Indexed 18.2K+0.2k
All runs Apr 14
Total 1,130
OpenAI 540
Anthropic 260
Google 180
Mistral 90
Others 60
800 600 400 200 0 Apr 11 Apr 12 Apr 13 Apr 14 Apr 15 Apr 16 Apr 17

Ask anything in plain English

Type a question. Zentu translates it into a trace lookup and answers with the receipts.

Learn more
Ask Zentu ⌘/
Q Why did support-triage cost more on Tuesday?
Three runs hit retries on the gpt-4o call after a tool timeout. Each retry replayed the full context — adding $18 across 14 minutes.
› Ask anything about your agents…

Drill into any trace

Every decision, tool call, and output — laid out as a timeline you can click through.

Learn more
Trace explorer · support-triage Success
run #4821a started 2m ago · 3.2s total
Tokens 3,420
Cost $0.018
Status 200 OK
Classify intent Routed to "billing-question" gpt-4o-mini 230ms
Search knowledge base 3 docs retrieved · vector search tool call 410ms
Generate reply & route Sent to tier-2 queue gpt-4o 2.1s
Ask anything

Questions Zentu can answer.

From "what did we spend?" to "is it worth it?" — in plain English.

Q. What did our agents cost this week, and was it worth it?
A. $312 across 1,240 runs — down 8% w/w. support-triage routed 412 tickets at $0.018 each — top value driver.
Q. Why did costs jump on Tuesday afternoon?
A. Three doc-writer runs hit retries after a tool timeout — adding $18 across 14 minutes.
Q. Which agents are actually earning their keep?
A. support-triage (412 tickets, 89% success) and sales-researcher (210 leads). Together: 78% of total value.
Q. Is the cheaper model giving us the same results?
A. Since the switch from gpt-4o to mini, success rate held at 89% — saved $148 this week alone.
Q. How many runs failed, and what was the most common reason?
A. 137 failures (11%). Top cause: tool timeouts (62) — clustered around the billing API between 2–4pm.
Q. What should I tell the board about our AI spend?
A. $312 spent, 1,240 runs handled, 89% success. ROI: ~17× vs. doing it manually. Trending the right direction.
How it works

Connect once. Then just ask.

Three steps. Works with any framework. Answers in plain English, out of the box.

1

Connect

Plug in your agents in minutes. OpenTelemetry-compatible. Works with LangChain, CrewAI, or your own framework.

$pip install zentu
›zentu.trace(my_agent)
Connected · 4 agents
2

Observe

Zentu captures every decision, tool call, token, and dollar — quietly, in the background. Nothing to maintain.

RUNS / DAY 1,240
3

Ask

Type a question in plain English. Zentu answers with the numbers and a sentence anyone in the room can understand.

Q Why did cost drop this week?
support-triage moved 60% of calls to mini — saved $148.

See what your AI agents are actually worth.

Join the waitlist. Free pilot for the first 10 teams.

30s Setup time
4 weeks To first ROI report
10 teams Free pilot spots