Skip to content

LLM observabilityKnow what every LLM call costs, and hear about it the minute one fails.

Alye records every model call your services make: cost by service, model and customer, latency, errors and retries. Two lines of code per service.

Get started freeRead the docs

Free for 500,000 calls a month. No card needed.

Overview · last 30 daysExample

Your 5 services made 1.24M calls that cost $17,099, 12% more than last month.

kb-ingest stopped reportingNo heartbeat for 42 minutes. On-call was paged.
Retires Dec 9gemini-2.5-pro3 services

Works with the providers and frameworks you already use

  • OpenAI
  • Azure OpenAI
  • Anthropic
  • Gemini
  • Vertex AI
  • Bedrock
  • DeepSeek
  • Groq
  • Mistral
  • OpenRouter
  • Python
  • Node.js
  • Express
  • NestJS
  • OpenTelemetry
01 · The invoice

One number from each provider. Now you see where it went.

The invoice says nothing about which service spent it, which model, or which customer it was for. Alye splits every dollar by service, model and customer.

Prices come from a versioned price list, with regional and marketplace prices for Azure, Vertex AI and Bedrock, and your own prices for fine-tunes.

How LLM cost tracking works

Spend by serviceExample
support-reply$8,940
ticket-triage$3,610
field-agent$2,880
kb-ingest$1,669

Priced per model and region, from a versioned price list.

02 · Quiet failures

See the calls that fail quietly.

A rate limit at 2 a.m. A retry that hides a timeout. Every request becomes a trace you can open step by step, with the model calls, retries and errors inside it.

Trace · answer_ticketExample
RateLimitError · 429Retried twice, then fell back
answer_ticket
retrieve_context
chat claude-sonnet
429
chat claude-sonnet
429
chat gpt-4.1-mini
03 · Setup

Two lines in each service.

Calls through the OpenAI, Anthropic and Google GenAI clients are recorded from then on. They go out in batches in the background, so your requests never wait.

Already on OpenTelemetry? Point your exporter at Alye instead.

# pip install alye
import alye
alye.init(service="support-reply")

// npm install alye
alye.init({ service: "support-reply" });
const openai = alye.wrap(new OpenAI());
04 · Alerts

Hear about it before your users do.

Spend limits, monthly forecasts, error rates, slow calls and services that stop reporting, checked every minute. Sent to email, Slack or Microsoft Teams.

AlertsExample
  • support-reply failed 7.4% of callsSlack · Last 15 minutes · over the 2% limitFiring
  • kb-ingest has not reported for 12 minutesTeams · Heartbeat missingFiring
  • This month is on track for $19,790Email · Budget $20,000Resolved
05 · Customers

Know what each customer costs you.

Tag each request with your customer ID to see spend per customer, set monthly budgets and open any customer's traces. Optional: everything works per service without it.

Customers · this monthExample
CustomerSpendBudget
northwind$2,410$3,000
contoso$1,190$2,000
fabrikam$640$1,500
06 · Privacy

Without reading your prompts.

Prompts and responses aren't sent unless you turn it on. When you do, emails, phone numbers, card numbers and keys are masked inside your service before anything leaves it.

Stored promptExample
User

Hi, I'm locked out. Reach me at [EMAIL] or [PHONE]. My card ending [CARD] was charged twice.

Masked inside your service by the SDK, before anything is sent. Off by default.

LLM observability questions

What is LLM observability?

LLM observability is recording every call your application makes to a large language model, with its cost, tokens, latency, errors and retries, so you can see what your AI features spend, find failing or slow calls, and get alerted before users notice. Alye does this with two lines of code per service.

How do I track OpenAI and Anthropic API costs per service and per customer?

Add the Alye SDK to each service and, optionally, tag requests with your customer ID. Alye prices every call from a versioned price list, including regional prices for Azure OpenAI, Vertex AI and Bedrock, and shows spend by service, model and customer.

Does Alye need a proxy or gateway in front of the model provider?

No. The SDK wraps the official OpenAI, Anthropic and Google GenAI clients inside your service and sends records in the background, so requests go straight to the provider and never wait on Alye.

Which providers and languages does Alye support?

OpenAI, Azure OpenAI, Anthropic, Google Gemini and Vertex AI, Amazon Bedrock, DeepSeek, Groq, Mistral and OpenRouter, from Python 3.10+ and Node.js 20+ with Express and NestJS integrations. Anything that already sends OpenTelemetry GenAI spans can send them to Alye.

Can Alye alert me when LLM costs or errors spike?

Yes. Alye checks spend, monthly forecast, error rate, p95 latency, call volume and services that stop reporting every minute, and sends alerts by email, Slack or Microsoft Teams.

Does Alye store my prompts and responses?

Not unless you turn it on. When you do, emails, phone numbers, card numbers and API keys are masked inside your service before anything is sent.

How much does Alye cost?

Free for 500,000 recorded calls a month. Team is $49 a month for 10 million calls, then $5 per 100,000, with unlimited members. Enterprise is custom, from $500 a month.

New to this? Read what LLM observability is and what to measure.

No surprises on the next invoice.

Create an account, add the SDK to one service and send a call. Your first trace shows up in about five minutes.

Get started freeSee pricing