delayed signal
Month-end surprise
Your LLM bill tells you what happened 28 days ago. By then, an unprofitable account has already completed a full billing cycle.
day 01$4.20 cost
day 28margin leak found
Building in public - Early access is free - no card, no expiry
Your AI bill shows spend. Tokonomics shows who's draining your margin.
Two customers on the same $49 plan. One costs $4 in tokens. One costs $51. Your provider dashboard shows $55 total. Tokonomics shows you which one to fix.
Customer Margin Overview
margin-ops.command / updated 43 seconds ago
LiveNo prompt capture
gross margin
86.4%
+4.2 pts from routing
monthly LLM COGS
$145.80
under budget
margin leaks
2
Two customers are below your configured margin floor.
| Customer ID | Monthly Revenue | LLM Cost | Gross Margin | Status |
|---|---|---|---|---|
| cust_2r5t | $49/mo | $44.10 | 10.0% | Unprofitable |
| cust_9x8q | $49/mo | $51.30 | -4.7% | Losing Money |
| cust_7p1n | $99/mo | $38.40 | 61.2% | Watch |
| cust_8a2f | $199/mo | $4.20 | 97.9% | Healthy |
| cust_3k9m | $99/mo | $6.80 | 93.1% | Healthy |
Detected automatically from customer usage + Stripe revenue.
The Problem
Your provider dashboard shows one number. It's the wrong one.
Total spend tells you what you burned. It doesn't tell you which customers burned it at a loss.
flat plan risk
Flat pricing, variable costs
Two customers pay the same $99. One costs $4 in tokens. One costs $51. Your provider dashboard treats them the same.
$99 plansame revenue
cust_a
$4 COGS
cust_b
$51 COGS
workflow spike
Runaway agent loops
Without workflow attribution, a runaway agent looks like normal usage until the invoice forces you to reverse-engineer it.
agent loop 1
agent loop 2
agent loop 3
unit economics
A profitable plan can still lose money.
Ten customers on a $49 plan can look like $490 MRR while two heavy users quietly push the cohort below gross margin.
2 heavy users$110 COGS
8 healthy users$40 COGS
$490 MRR can still be negative margin
the missing layer
Spend only matters when it is tied to revenue.
Tokonomics connects provider spend to customer, feature, workflow, and Stripe revenue, so the margin story is visible before the invoice.
provider.bill$5,000
customer.margin?
feature.cogs?
AI cost reckoning
Big teams found out the hard way.You don't have to.
$500M. One month. One client.
An enterprise client was reportedly billed $500M in one month — no per-customer visibility, no alerts. The invoice was the first signal.
support copilot startup
Found 3 customer accounts below a 20% margin floor in the first week - none of them visible in the provider dashboard.
Action taken
Set margin floor alerts on all three. Repriced two at next renewal.
writing app, ~300 customers
Discovered one feature route (document rewrite) was responsible for 41% of total token spend - called by 8% of users.
Action taken
Moved that route to a cheaper model. Margin improved by 11 points within 30 days.
ai search tool, seed stage
Identified that free-tier users were consuming 3× more tokens per session than paid users. The free tier was loss-making at scale.
Action taken
Added a token cap to free tier. Conversion rate held. LLM COGS dropped 22%.
The same visibility gap that cost enterprise teams millions happens quietly inside early-stage AI products. Tokonomics was built to catch it before month-end.
How It Works
Start with spend visibility. Upgrade into real margin control.
Tokonomics gives you value at every setup step, from zero-code provider visibility to customer-level gross margin.
provider keyread only
sk-ant-****************
total spend
model mix
spike alerts
01
Paste your API key. See your spend in 30 seconds.
Read-only key. Zero deploy. Paste your OpenAI, Anthropic, or Gemini key and you'll immediately see total spend, model mix, cost trends, and spike alerts. No code. No infrastructure change. Just visibility.
Read-only keyZero deployNo code required
visible in 30 seconds
paste into Cursor, Replit, Lovable, or Claude Code
Add tokonomics tracking to every LLM call. Use the logged-in user as customer_id, infer the feature name from the route, and keep prompt content private.
02
Add one line. Get cost per customer by tomorrow.
Tell Tokonomics your stack - Cursor, Replit, Lovable, Claude Code, or hand-coded - and you get a ready-made snippet. Drop it in and every LLM call gets tagged to the customer and feature behind it. One line. Real attribution.
No prompt capturedWorks with your existing stack
one line of tracking
cust_8a2f97.9% margin
$199 revenue$4.20 AI cost
cust_2r5t-4.7% margin
$49 revenue$51.30 AI cost
03
Connect Stripe. Identify unprofitable accounts this week.
Revenue minus AI cost, per customer, updated daily. That's the number you've been missing - and the one your investors will eventually ask about.
margin per customer, daily
time to value
Day 0
Total AI spend, model mix, spike alerts
Day 1
Cost per customer, feature attribution
Week 1
Gross margin per customer, first margin leaks flagged
Security
What Tokonomics can and cannot see
You can revoke access in seconds. We recommend rotating your key after setup as standard practice.
API usage metadata
We see model name, token count, timestamp, and cost. That is all.
No prompt content
We never capture, log, or store what you or your users send to the model.
No completion content
We never capture model responses. Only the cost and metadata of each call.
Read-only provider access
Your API key is used to pull usage data only. We cannot make API calls on your behalf.
Features
Detectwhat's leaking. Diagnose why. Fix it before next month.
You're not running a DevOps team. You don't need another observability tool. You need to know which customers are costing you money - and what to do about it.
detect
Start here ->Per-customer attribution
Every LLM call tagged to a customer, feature, and workflow. See your exact cost-to-serve - per customer, live. No aggregates. No guessing. Just the number.
customer.marginbreached floor
cust_8a2f97.9%
$4.20Healthy
cust_7p1n61.2%
$38.40Watch
cust_9x8q-4.7%
$51.30Losing
AI gross margin dashboard
Stripe revenue minus LLM COGS, per customer, updated daily. The number your investors will ask about - and the one that tells you who's actually making you money.
gross.margin
86.4%+4.2 pts
COGS
$145.80
Budget alerts & circuit breakers
Set a daily or monthly cost ceiling per customer or feature. Get the alert before it becomes a crisis - not after the invoice confirms it.
daily cap82% used
Alert before customer spend crosses $42 today.
diagnose
Surface the score, weekly signal, and routes worth switching before deciding what to change.AI Margin Score
One number that tells you whether your AI unit economics are improving - and exactly what's dragging the score down this week.
AI Margin Score
78
heavy usersdrag -12
model wastedrag -7
pricing fitlift +4
Weekly Margin Digest
Every Monday in Slack or email: which customers crossed your margin floor, what shifted, and 1-2 actions worth taking. Short enough to read before standup.
Monday digestSlack + email
2 customers crossed margin floor
1 workflow spiked 240%
3 model swaps ready for review
fix
Model routing suggestions
Think Haiku could replace GPT-4o for your simpler queries? Test it on real traffic first. Side-by-side quality comparison - so you know before you ship.
before
GPT-4o
$0.038/call
after
Haiku
save $38/mo
quality shadow testsafe
Usage-based pricing simulator
Wondering if usage-based pricing would fix your margin problem - or kill your conversions? Simulate the MRR impact before you touch your pricing page.
pricing.modeusage based
Fair-use limit120K tokens
MRR lift
+4.1%
margin
91.2%
Integrations
Works with the LLM frameworks you already use.
# Install: pip install tokonomics
from tokonomics import track, Tokonomics
toko = Tokonomics(api_key="tok_live_...")
@track(customer_id=request.user_id, feature="chat")
async def generate_response(message: str) -> str:
response = await openai.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": message}]
)
return response.choices[0].message.contentFounder pains
What founders use Tokonomics for
cust_8a2f97.9%
cust_7p1n61.2%
cust_9x8q-4.7%
Hidden heavy users
cust_9x8q
"One customer on my $49/mo plan costs me $51 in tokens every month. I had no idea until I checked manually. I need to see this without having to dig."
→ Found automatically. Flagged every time it recurs.
-4.7% marginflat plansame price
Usage bucket 1
Usage bucket 2
Usage bucket 3
Pricing uncertainty
pricing.sim
"I think I need to move to usage-based pricing. But I don't know if that means $400 more in MRR or losing 20% of my signups."
→ Model the change before touching your pricing page.
+$420 MRRGPT-4o$0.038/call
Haikusave $38/mo
Model waste
route.waste
"I'm 90% sure my app is calling GPT-4o on tasks that Haiku could handle. I just can't prove it without spending a full day in logs."
→ Prove it in 2 minutes. Ship the swap safely.
$38/mo leakapp route?
model calledunknown
monthly riskinvisible
Vibe-coded stack
vibe.build
"I built this fast. I genuinely don't know what model my app calls, how often, or at what cost. I need the map before I touch anything."
→ Map every LLM call by customer, route, and feature.
unknown AI stackROI Calculator
What would better margin visibility actually be worth to you?
Move the sliders. See how quickly margin clarity pays for itself.
benefit.model
routing25% lower COGS
repricing+4% revenue lift
incidentsbudget guardrails
directional estimate, tuned for early AI SaaS teams.
Estimate your monthly upside.
Move the inputs to see how quickly margin visibility pays back.
estimated.roi
10.8x
Estimated monthly benefit
Model routing$1,250 /mo
Repricing lift$396 /mo
Incident prevention$500 /mo
Total benefit$2,146 /mo
Growth plan$199 /mo
Net gain$1,947 /mo
How we calculate this ↓
Model routing savings: We assume routing 25% of calls to a cheaper model reduces COGS by 25% on those calls, based on typical GPT-4o → Haiku cost ratios.
Repricing lift: We estimate 4% additional revenue from repricing accounts identified as under the margin floor. Actual lift depends on your pricing elasticity.
Incident prevention: We estimate one usage spike per quarter at 30% of monthly spend, preventable with budget alerts. Divided across 12 months.
These are directional estimates, not guarantees. Your actual results depend on your stack, customer mix, and pricing model.
If Tokonomics helps you catch one unprofitable customer or one wasteful workflow, it has likely paid for itself.
Directional estimate - real results depend on your stack, pricing model, and usage patterns. But the math tends to hold.
Pricing
Start free. Pay only when you go deeper.
You don't need to integrate everything on day one. Start with your API key, get real spend data, and upgrade when the next layer makes sense.
Free
$0/month
See your spend. Immediately.
- Best for
- Founders who want their first honest look at where the money is going.
- Note
- No card. No expiry. No catch.
payback: Worth it if it gives you your first honest spend view before the next invoice.
- Connect your LLM provider key
- Total AI spend and trends
- Model mix and provider breakdown
- Spend-spike alerts
- Read-only setup
- Free forever
Starter
$99/month
See what each customer actually costs you.
- Best for
- Teams connecting AI spend to individual customers for the first time.
- Note
- 90-day customer cost history
payback: Worth it if it catches one bad account this month.
- Everything in Free
- Per-customer cost attribution
- Per-customer cost table
- Feature and workflow tagging
- Alerts per customer
- Cursor/Replit prompt assist
Growth
Maximum Value
$199/month
See real margin. Then fix what's breaking it.
- Best for
- Revenue-connected founders closing the gap between spend and profit.
- Note
- Unlimited cost and margin history
payback: Worth it if one routing swap or reprice improves margin by 3–4 points.
- Everything in Starter
- Stripe integration for gross margin
- Pricing simulator
- Safe model routing tests
- Weekly margin digest
- AI Margin Score
- Slack and email notifications
Scale
$399/month
Margin ops across the whole team.
- Best for
- Teams who need shared visibility, investor reports, and margin automation.
- Note
- Workspace + API access
payback: Worth it if multiple people need one shared margin source of truth.
- Everything in Growth
- Team access
- Benchmark comparisons
- White-label investor reports
- API access
- Margin copilot drafts fixes
- SLA and DPA available
Free during beta for all early access users. No card required to start.
FAQ
Common questions
How is this different from Helicone, Langfuse, or other LLM observability tools?+
Helicone and Langfuse are built for debugging - they track prompts, latency, and errors so engineers can improve model performance. Tokonomics is built for unit economics - it tracks cost per customer, connects it to revenue, and tells you which customers are making you unprofitable. Different job. If you already use an observability tool, Tokonomics sits alongside it, not instead of it.
Do I need Stripe connected to get value?+
No. Step 1 - paste your provider key - gives you real spend data with zero code. Step 2 - add one tracking line - gives you cost per customer. Stripe is Step 3 and unlocks gross margin per customer. You get meaningful value before you ever touch Stripe.
Does setup require code changes to my app?+
Step 1 requires zero code - you paste a read-only API key and see your spend in 30 seconds. Step 2 requires one line of tracking code added to your LLM calls. Step 3 (Stripe) is a connection, not a code change. Most founders are fully set up within an hour.
What exactly does Tokonomics store about my customers?+
We store usage metadata - model name, token count, cost, timestamp, and the customer ID you assign. We never capture prompt content or model responses. We have no visibility into what your users send to the model or what the model returns. Full details are in our privacy block above.
What LLM providers and frameworks are supported?+
OpenAI, Anthropic, and Google Gemini natively. LangChain and LlamaIndex via wrapper. If you're building in Cursor, Replit, or Lovable, we have ready-made setup snippets for those environments. If you're using a provider not listed here, reach out - we're adding integrations fast.
What happens to my data if I cancel?+
Your data is available to export before cancellation. After cancellation, your account and associated data are deleted within 30 days. You can remove our access by revoking your API key at any time - no dashboard action required.
More questions? Write to us at hello@tokonomics.dev
Before you ship more AI features, make sure your best customers are actually profitable.
Start with total AI spend visibility - no card, no code, no deploy. Add customer attribution and Stripe margin when you're ready.
Zero prompt captureRead-only API keyFree during beta