C
CARTIEAI
PricingTry free
Back to home
Proprietary engine

CARTIE Lens

The first cost linter for application code. SonarQube lints code quality. Snyk lints security. We lint cost.

Paste Python or JavaScript → see every cost-expensive pattern (LLM-in-loop, missing pagination, abandoned streams, retry storms) with the projected dollar-impact and a one-line fix.

Paste your code
Up to 50,000 characters. Nothing is stored unless you sign in and click "Scan & Save".
Findings
Each finding has a fix + projected $-impact at observed traffic.
Paste code on the left, hit Scan.

The catalog

10 cost-pattern signatures live today. Each one is a real anti-pattern we've seen in production codebases — and a documented $-impact at observed traffic.

L001
python
high
LLM call inside for/while loop without batching
OpenAI/Anthropic calls inside a loop pay per-iteration latency tax + miss batch-API 50% discount.
L002
python
high
Cloud SDK list call without pagination
boto3 / google-cloud / azure-sdk calls without paginator silently truncate or repeatedly refetch.
L003
python
medium
Streaming LLM response not fully consumed
Streaming = you still pay for every output token generated, even if your code never reads them.
L004
python
medium
Stable system prompt without prompt-cache marker
Long system prompts repeated across requests are 90% cheaper if marked with provider cache_control.
L005
python
high
Retry decorator without exponential backoff
Naive retry decorators (@retry(tries=5)) trigger retry-storms during provider degradation — you pay for every failed attempt.
L006
javascript
high
fetch() inside for/forEach without Promise.all batching
Sequential awaits inside a loop serialise N requests instead of parallelising.
L007
javascript
medium
OpenAI/Anthropic SDK call in tight loop without rate-limit
Rapid-fire LLM calls hit per-minute rate limits, triggering 429 retries that you pay for.
L008
javascript
low
console.log inside high-throughput request handler
CloudWatch / Stackdriver ingest costs scale per GB of log volume.
L009
python
medium
ORM query inside a loop (N+1 query antipattern)
Each iteration issues a separate DB query — N queries instead of 1.
L010
python
low
time.sleep() inside async coroutine
Blocking sleep in async code freezes the event loop — you pay for idle compute on every request.
Coming next

PR comments + VSCode squiggles + standalone CLI.

Cost-Lens will run as a GitHub Action (15 lines of YAML) and comment on every PR. A VSCode extension surfaces findings as inline squiggles while you type. Standalone CLI binary for local pre-commit.

Want early access? Join the waitlist at /vscode-extension

We value your privacy. Cookies help us improve your experience. Learn more

Install CARTIE AI

Add to your home screen for quick access and offline support