New

The first monitoring built for LLM APIs.

A plain 200 OK means nothing for an LLM: the canary measures the real time to first token, and validates both the content and format of the response — not just the HTTP status.

canary checks · llm
api.anthropic.com/v1/messages claude-haiku-4-5 UP412ms
api.openai.com/v1/chat/completions gpt-4o-mini UP380ms
bedrock-runtime.eu-west-1 claude-3-haiku DEGRADED2.1s
api.mistral.ai/v1 mistral-small RATELIMIT429

A binary state isn't enough for an LLM.

Most LLM incidents still return 200 — with a token that never arrives, incorrect content, or malformed JSON. That's exactly the state classic uptime monitoring can't see.

UP

HTTP 200, TTFT under threshold, content and format valid.

DEGRADED

200 but TTFT exceeded, incorrect content, or invalid format — the most common case in practice.

DOWN

Non-200, timeout, connection refused, interrupted stream.

RATELIMIT

An isolated 429: quota/backoff remediation, not a provider incident.

How it works.

01

Deterministic canary

Temperature 0, trivial expected reply ("PONG") — near-zero cost even at 1 check per minute.

02

Timing the real first token

TTFT is measured at the first text chunk, not the first HTTP byte — otherwise the degradation goes unnoticed.

03

Content + format assertions

A "PONG" canary and a JSON schema canary also catch silent structured-output regressions.

t0 · request sent TCP/TLS established HTTP headers received (200) first token ← TTFT stream ended
Pitfall #1: HTTP headers (grey zone) arrive well before the first token (start of the green zone) — a check that times too early masks exactly the degradation it's meant to catch.

What's included.

Real TTFT

Streaming time-to-first-token, threshold calibrated per model — never a global threshold.

Content canary

The expected reply is validated word for word — catches a response silently drifting off.

Format canary

A second check validates a JSON schema — catches structured-output regressions after a model update.

Compatible with your AI stack.

Three technical families cover the market — an OpenAI-compatible parser alone covers over two-thirds of providers, self-hosted included. ★ = reference format, implemented natively; the others build on one of the three.

★ Anthropic ★ OpenAI ★ AWS Bedrock Azure OpenAI Mistral Groq Together AI Fireworks OpenRouter DeepSeek xAI (Grok) Google Gemini Cohere Ollama vLLM / TGI self-hosted

Anthropic, OpenAI and AWS Bedrock cover the market's three stream formats (typed SSE events, SSE chunks, binary event stream) — every other provider listed builds on one of these three parsers.