Free
$0 / mo
Try the proxy on a monthly allowance.
150
requests included · hard cap, no overage
LLM FinOps proxy for OpenAI and Anthropic
Luspend is a hosted meter for European platform teams. Point the SDK at one URL, keep your provider keys, and track spend by team and feature. Prompts are not stored. The dashboard runs in Frankfurt.
What it is
A meter on the path to the model
Your app talks to Luspend. Luspend talks to OpenAI or Anthropic. You see who spent what, without a spreadsheet and without a prompt warehouse.
- 01
Keep your keys
OpenAI and Anthropic stay yours. We never hold them.
- 02
Change one URL
Point the SDK at Luspend. The rest of the call stays the same.
- 03
See the spend
Cost, tokens, team, feature. Not the prompt.
Your app↓→Luspend↓→OpenAI / Anthropic
For EU infra
Built for people who own the LLM path
Luspend is for developers and technical owners who already run production model calls and need FinOps without a spreadsheet or a prompt warehouse.
Platform and infrastructure
Drop a proxy in front of OpenAI and Anthropic without owning another gateway cluster. Tokens, cost, and tags show up in one dashboard.
Engineering leads
Give product teams a monthly request allowance. See which feature is burning the bill before finance asks.
Security and compliance
Provider keys never land in Luspend storage. Prompts and completions are not retained. The control plane sits in Frankfurt.
- Providers
- OpenAI and Anthropic SDKs, same request shape
- Auth model
- Your provider key plus an org proxy token
- What we store
- Usage, cost, tags. Not prompts.
- Dashboard region
- Vercel Frankfurt (fra1)
- Billing unit
- Proxy requests, not a token markup
- Audience
- EU product teams with in-house LLM traffic
Latency
One extra hop. Less than a quarter second.
Measured as time to first token on a streamed ~140-token reply. Direct to the provider was 554 ms. Via Luspend was 738 ms. Median extra hop was 184 ms. Your numbers will follow your provider region and payload.
Direct554 ms
Via Luspend738 ms
Extra
+184
ms median
- Tiny JSON+206 ms
- Tiny stream+269 ms
- Medium JSON+254 ms
- Stream first token+248 ms
Setup
Four lines. Same SDK.
Keep OPENAI_API_KEY or ANTHROPIC_API_KEY. Set baseURL to api.luspend.com. Send x-proxy-token. Tag team and feature if you want allocation.
- 1
Keep the provider key
Same secret you already use. Luspend never stores it.
- 2
Point at Luspend
OpenAI uses /v1. Anthropic uses the origin. SDKs append the rest.
- 3
Send your Luspend token
Created in the dashboard. Goes on x-proxy-token.
- 4
Tag the spend
Optional. Team and feature show up on every request.
import OpenAI from "openai"; const openai = new OpenAI({1 apiKey: process.env.OPENAI_API_KEY,2 baseURL: "https://api.luspend.com/v1", defaultHeaders: {3 "x-proxy-token": process.env.LUSPEND_TOKEN,4 "x-proxy-tag-team": "search",4 "x-proxy-tag-feature": "chat", },});Compare
A meter, not another prompt store
If you need payload replay or a self-hosted mesh, pick a gateway. If you need spend by team on the calls you already make, use Luspend.
| Product | Model | Provider keys | Prompts | Best fit |
|---|---|---|---|---|
| Luspend | Hosted meter | Stay in your app | Not stored | Cost allocation for OpenAI and Anthropic |
| LiteLLM | Usually self-hosted | You operate the gateway | Your config | Multi-provider routing you maintain |
| Helicone | Hosted observability | Gateway or async logging | Often retained | Traces, prompts, eval-style debugging |
Pricing
Pay for the meter. Not the tokens.
OpenAI and Anthropic stay on your bill. Luspend charges USD for the proxy meter: a monthly floor plus the requests you send. Free is 150 requests per calendar month with a hard cap.
Start free
Solo
$19 / mo
One product. One team.
25k
requests included · $0.50 / 1k after that
Team
Most teams$49 / mo
The default for a working shop.
150k
requests included · $0.25 / 1k after that
FAQ
Direct answers for infra owners
Short facts you can paste into an architecture review, a DPA questionnaire, or an LLM answer.