Real-time LLM cost intelligence
Your AI spike happened
three days ago.
Your dashboard just told you.
LLMtrack shows every dollar the instant a request completes — broken down by feature, model, and user. One async call. No proxy. No latency added to your app. Official Node & Python SDKs.
Free forever planData in under 1 secondNo credit cardPrompts never stored
Requests
48
↑ 12%
Total cost
$0.23
↑ 8%
Tokens
48.3K
→ stable
Error rate
2.1%
↑ alert
FeatureRequestsCost
chat-completion512$0.0031
summarizer198$0.0008
search-assist137$0.0003
Tracking costs across every major LLM provider
OpenAIAnthropicGoogle GeminiMistralGroqCohere
<1ms
Added latency to your app
1
Line of code to integrate
6
LLM providers supported
<1s
Time to see your first cost data
The problem
Your AI bill is a monthly mystery.
Provider dashboards show you a total, once a day at best. By the time you see the spike, it's already cost you.
Surprise bills at month-end
A single agent loop with bad context management can cost 10× your normal daily average. You find out when Stripe charges you — not when the loop runs.
Real case: $200 → $6,000 in one month
No feature-level visibility
Your provider dashboard shows a single monthly total. Not which chatbot, summarizer, or search feature is responsible. You're optimizing blind.
Average: 3 features, zero cost breakdown
Model switching in the dark
A budget tier like GPT-5.6 Luna or Gemini 3.6 Flash can be 80–95% cheaper for your specific use case. But without your real token counts per request, you're guessing at benchmarks.
Companies overpay 50–90% on wrong models
How it works
Running in under 3 minutes.
No proxy. No configuration. No new infrastructure. Install the SDK and add one line after your existing LLM calls.
01
Create workspace
Sign up free in 60 seconds
No credit card. No sales call. Just your email and a workspace name. Your ingestion key is generated instantly — set it once as LLMTRACK_API_KEY in your environment.
Your API key
llmt_k_••••••••••••••••
✓ Generated instantly
02
Install the SDK
One line after your LLM response
Fire-and-forget, never throws, and adds zero latency to your app.
npm install llmtrack · Python: pip install llmtrack-sdk · Open source on GitHub → · Prefer raw HTTP? →import { LLMtrack } from 'llmtrack';
const llmtrack = new LLMtrack({ apiKey: process.env.LLMTRACK_API_KEY });
// after your LLM call — fire-and-forget, never throws, never blocks
llmtrack.track({
provider: 'openai',
model: 'gpt-5.6-sol',
feature: 'chat-completion',
promptTokens: 1200,
completionTokens: 480,
reasoningTokens: 3100,
});03
See costs instantly
Your dashboard populates in under 1 second
Cost by feature, model savings, spike alerts, and spend forecast — all from that one call. Not 24–48 hours later like your provider.
Cost per request$0.0031
Chat-completion share64%
Monthly model savings$7.28
Days until budget exceeded22 days
What you get
Built for cost clarity, not complexity.
Real-time tracking
See costs the moment they happen
Provider dashboards update every 24–48 hours. LLMtrack reflects data in under 1 second. You know about a spike before it compounds.
Data visible in under 1 second after every request
Cost spike alerts fire in real-time — not end-of-day
Compare cost across the same feature hour-by-hour
just nowchat-completion$0.00022391ms
2s agosummarizer$0.028802,440ms
8s agosearch-assist$0.002181,368ms
14s agocode-review$0.040503,488ms
Feature analytics
Which feature is eating your budget?
Every tracked request carries a feature_name you define. LLMtrack groups and ranks them — so you instantly see your chatbot costs 4× more per request than your summarizer.
Cost share % per feature — ranked by biggest spend
Week-over-week trend — catch degradation early
Error rate per feature — surface costly failures
| Feature | Requests | Cost | Trend |
|---|---|---|---|
| chat-completion | 198 | $0.0031 | ↑ 12% |
| summarizer | 137 | $0.0008 | ↓ 3% |
| search-assist | 89 | $0.0003 | → 0% |
| code-review | 62 | $0.0002 | ↑ 4% |
Model intelligence
Stop overpaying for the wrong model
LLMtrack calculates what your usage would cost on every cheaper alternative — using your real average token counts per request, not benchmarks. Savings ranked by impact, biggest first.
Calculated from your actual token distribution
All alternatives ranked by monthly savings
Direct link to provider docs for each alternative
gpt-5.6-sol → gpt-5.6-luna+99%$8.96
gemini-3.1-pro → gemini-3.6-flash+91%$0.03
claude-opus-5 → claude-haiku-4-5+35%$0.01
Spend forecast
Know your bill 7 weeks before it arrives
LLMtrack fits a linear regression to your last 14 days of usage and projects forward through the end of next month. Know before the invoice hits.
Projected this month and next — updated daily
Budget crossing point shown as days remaining
Email and Slack alerts fire before you overspend
This month
$8.07
Trend: stable
Next month
$26.76
Based on daily usage
Budget exceeded
22 days
$20/mo budget
⚡ Switch & save
$8.96/mo
→ gemini-3.6-flash
What developers are finding
"I had no idea that Google's models perform on par with OpenAI for most of what I build — and cost a fraction of the price. Seeing the real numbers side-by-side in the dashboard was genuinely surprising."
"Tracking requests the moment they happen changed how I think about my app. I used to wait until the next day hoping I hadn't overspent. Now I see exactly what's happening in real time — the way it should have always worked."
"I can finally see what my API is actually costing me, broken down properly. Not a monthly total from a provider dashboard — real data, by feature, as it happens."
It started the way most of these stories start: with a number that didn't make sense.
Why we built LLMtrack →How we compare
Built for builders. Not enterprises.
Helicone and Langfuse are powerful — and built for dedicated ML ops teams with weeks to integrate. LLMtrack is for the developer who needs answers this afternoon.
| LLMtrack | Helicone | Langfuse | LangSmith | |
|---|---|---|---|---|
| Entry price | $0 forever | $0 (10k/mo) | $0 (50k units) | $0 (LangChain) |
| Proxy required | ✓ No proxy | Partial | ✓ No proxy | SDK lock-in |
| Official SDKs | ✓ Node + Python (MIT) | ✓ | ✓ | Required |
| Setup time | <5 minutes | ~15 min | ~30 min | 60+ min |
| Real-time data (<1s) | ✓ Yes | ✗ ~minutes | ✗ ~minutes | ✗ ~minutes |
| Feature-level cost | ✓ Built-in | Partial | Via traces | Partial |
| Business unit economics | ✓ Cost per user | ✗ | ✗ | ✗ |
| Spend forecast | ✓ All plans | ✗ Enterprise | ✗ | ✗ |
| Model savings calc | ✓ Automatic | ✗ | ✗ | ✗ |
| Pro plan price | $11.99/mo | $120/mo | $29/mo | $39/seat/mo |
Pricing
Simple pricing. No surprises.
Start free. Pay only when your usage grows. No annual lock-in. No sales calls.
Free
The full experience, free
$0
forever
3,000 events / month
Every feature unlocked
Unlimited integrations
Official Node & Python SDKs
Model savings recommendations
Feature & per-customer cost breakdown
Spend forecast
CSV export
30-day data retention
Best value
Pro
More volume and longer history
$11.99
per month
Everything in Free
25,000 events / month
Official Node & Python SDKs
Full analytics dashboard
Email alerts
Model savings recommendations
Feature cost breakdown
Spend forecast
Business metrics (cost per user)
CSV export
90-day data retention
12 months data retention — $16.99/mo
Toggle for 12 months if you need longer-term history
Business
For SaaS teams with multiple models and stakeholders
$49
per month
100,000 events / month
Everything in Pro
Official Node & Python SDKs
Per-customer cost tracking NEW
Budget limits per key & feature NEW
Email + Slack + Webhook alerts
PDF finance reports
5 team seats
1-year data retention
We count requests, not traces or spans. 3,000 events is 3,000 real API calls.
FAQ
Questions we get a lot.
No. The SDK's track() is fire-and-forget and never throws. Raw-HTTP users add .catch(()=>{}) themselves. Tracking runs after your LLM response and adds zero latency to your app.
Your provider's usage dashboard is updated on their schedule — typically once per day. LLMtrack is different: when your app calls our ingest endpoint, we immediately write that record to our database. Your dashboard reads directly from that database in real time. We're not polling your provider — we're receiving your data directly.
Never. LLMtrack stores only metadata: token counts, computed cost, model name, feature name, latency, and status code. Your actual prompts and generated responses never touch our infrastructure. Row-level security at the database layer ensures complete workspace isolation.
LLMtrack maintains a server-side pricing table covering current models from OpenAI, Anthropic, Google, Mistral, Groq, and Cohere — including reasoning-token and cached-token rates. On each ingest we compute cost from your token counts against that table. Cost is always computed server-side; a client-supplied cost value is ignored. If we don't have pricing for a model yet, the response says pricing_status: "unknown_model", so a $0 cost is never silently wrong.
Yes. Node uses the ESM-only npm package
llmtrack on Node 18+, and Python uses the PyPI package llmtrack-sdk, imported as llmtrack_sdk. Both provide automatic idempotency, retries, and track() that never throws. See the SDK guide.Yes. Reasoning and thinking tokens are billed at output rates by providers. LLMtrack accepts
reasoning_tokens and prices them. Read the reasoning-tokens guide.Three things. First: real-time data in under 1 second vs. delayed aggregation. Second: business unit economics — cost per user, break-even analysis, and margin impact — which neither competitor offers at any price tier. Third: automated model savings recommendations calculated from your actual token distribution, not generic benchmarks.
Stop guessing.
Start knowing.
Free forever plan. One line of code. Your first cost breakdown in under 5 minutes.
Free forever planNo credit cardNo proxy requiredData in under 1 second