Connect

Drop in our proxy SDK or point your gateway at our endpoint. Zero code rewrites.

Avg 2 minutes

Analyze

We trace every request — provider, model, prompt fingerprint, tokens, latency, retries.

Avg 4 minutes

Optimize

Get prioritized recommendations and turn on routing rules with one click. Roll back instantly.

Avg 3 minutes

Connect your app

import OpenAI from "openai";

const openai = new OpenAI({
  baseURL: "https://api.spendtensor.io/v1",  // ← one line
  apiKey: process.env.OPENAI_API_KEY,
});

// Every request is now traced, routed, and optimized.
await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Save me money." }],
});

Step 02 — Analyze

What we capture on every call

Metadata only by default. Full prompt and completion capture is opt-in per workspace, with redaction and sampling rules you control.

provider / modelNormalizes cost across OpenAI, Anthropic, Gemini, Azure, and Bedrock

input & output tokensExact per-call cost using live provider rate cards

prompt fingerprintFinds duplicate work that caching can eliminate

latency & retriesSurfaces silent retry storms that double your bill

team / app / env tagsChargeback and showback without spreadsheets

cache & route decisionProves realized savings for every optimization

Traced AI request metrics in the SpendTensor dashboard

Model routing decisions with quality gates

Step 03 — Optimize

Turn savings on, one rule at a time

Recommendations are ranked by projected monthly savings and implementation risk. Each one shows the evidence behind it — traffic volume, current model, candidate model, and eval scores.

Semantic caching for repeated prompts
Model downgrades that clear your quality floor
Prompt compression and system-prompt dedupe
Batch windows for non-interactive workloads
Retry and timeout tuning to kill silent waste

A typical first month

Day 0

Connect and baseline

Point one service at SpendTensor or connect read-only provider keys. We backfill up to 90 days of historical usage so you start with a real baseline, not an empty chart.

Day 1–3

Attribution and budgets

Tag traffic by team, app, and environment. Set budgets with 50/80/95% alerts so nothing surprises you again.

Week 1

First optimizations

Enable semantic caching and the top-ranked model swaps. Most teams see a measurable drop in daily spend within five business days.

Week 2–4

Routing at scale

Ramp routing rules across services with quality gates and rollback. Finance gets a verified savings report at month end.

Two ways to connect

Proxy mode instruments every request in real time. Read-only key mode pulls provider usage APIs on an hourly schedule — no traffic changes at all.

Start with one service

Roll out per service or per environment. Staging first is a common path; the dashboard separates environments cleanly.

Nothing risky by default

Routing and caching are opt-in. Until you enable them, SpendTensor only observes — zero behavior change to your product.

Hourly sync, live proxy

Proxy traffic appears within seconds. Key-based ingestion syncs hourly and reconciles against provider invoices.

See it running on real data

Open the interactive demo dashboard, or have our team walk your stack and estimate your savings.

Open live demo Book live demo View pricing

From signup to savings in 9 minutes.