How Agent Status Works

The Big Picture

Your Agent ←── Agent Status Validation ←── Real Device in Tokyo
           ←── Agent Status Validation ←── Real Device in London
           ←── Agent Status Validation ←── Real Device in New York
           ←── Agent Status Validation ←── Real Device in Lagos

Every check sends real requests to your agent from multiple geographic locations, measuring:

• Reachability — Can we connect at all?
• Latency — How fast does it respond?
• Correctness — Is the response valid and sensible?

How a Check Works

Dispatch — Agent Status selects nodes matching your criteria

Validation — Each node sends your configured prompts

Validate — Responses checked via evaluation prompts and schema validation

Aggregate — Results combined into a single verdict

Alert — If status changes, webhooks fire and emails send

The Validation Stack

Layer 1: Reachability

• Can we reach the endpoint?
• Does it return HTTP 200?
• Does it respond within the timeout?

Layer 2: Schema

• Is the response valid JSON?
• Does it match the expected format?
• Are required fields present?

Layer 3: Correctness (Evaluation Prompts)

Deterministic prompts with known-correct answers:

• "What is 2+2?" — Must contain "4"
• "Return JSON: {x:1}" — Must be valid JSON
• "Echo: hello" — Must contain "hello"

Layer 4: Latency SLA

• TTFB — Time to first byte
• Total Latency — Full response time
• P50/P95/P99 — Percentile tracking

Verdicts

Verdict	Meaning
UP	≥95% reachability, ≥80% eval pass, SLA pass
DEGRADED	Partial failures or SLA miss
DOWN	<20% reachability or majority failures

Why Real Devices?

Synthetic checks from data centers miss real-world issues:

• ISP routing problems
• CDN edge issues
• Geographic DNS failures
• Mobile network latency

Agent Status uses actual devices running in homes and offices worldwide.

What We Don't Do

• We don't store your responses — Only metadata
• We don't train on your data — Responses are validated and discarded
• We don't access your system — We only call the endpoint you give us