Home › Observability
Observability

Every AI request,
completely visible.

Full-stack observability for your AI layer. Trace requests end-to-end, understand cost per call, catch latency regressions before users do.

Start free
Live Request Trace 4,218 req / min
Request
POST
/v1/chat/completions
Explane
Routing
8ms · 4 policies
Provider
gpt-4o
OpenAI · tier-2
Latency
187ms
p99: 312ms
Cost
$0.0024
1,248 tokens
Response
200 OK
success · cached
4,218 req / min
31ms avg latency
99.99% uptime
0.04% error rate
What we capture

Complete signal,
zero blind spots.

Every dimension of your AI traffic — latency, cost, token counts, provider health, routing decisions, and a full audit trail — captured automatically, no instrumentation required.

Latency · p50 / p95 / p99
31ms p50
p95: 89ms · p99: 312ms
↓ 12ms from last week
Token Usage
2.4B
tokens · this month
↑ 18% growth
Total Cost
$1,840
this month · all providers
↑ 6% vs last month
Provider Health
3 / 3
providers operational
Last incident: 14 days ago
Model Mix
gpt-4o
top model · 62% of traffic
+claude-3.5 · +gemini-2.0
Error Rate
0.04%
429s + 5xx this hour
↓ 0.02% vs baseline
Routing Decisions
98.2%
policy match rate
4 active rules
12:34:01 sarah@acme.co Updated budget policy · Engineering team limit $1,000 → $1,500 governance
11:52:44 api:ex_k9a2 Failover triggered · azure-gpt-4 503 → gemini-2.0-flash routing
09:14:22 james@acme.co Added provider · Mistral Large 2 · embedding + chat endpoints integrations
Request Timeline

Distributed tracing,
built for AI.

Every request generates a complete waterfall trace — provider call time, token counting, policy evaluation, streaming overhead. Find exactly where your latency is going.

  • Span-level breakdown: routing, inference, streaming, post-processing
  • Compare traces across providers to surface performance regressions
  • Export to Datadog, Grafana, or any OpenTelemetry-compatible backend
trace_9f2b · req_8h74k 264ms total
0ms64ms128ms192ms264ms
sdk.request
264ms
policy.eval
14ms
route.select
8ms
gpt-4o.call
156ms
stream.out
76ms
token.count
8ms
Total wall time 264ms
Cost Intelligence

Know exactly what
you're spending, and why.

Real-time cost attribution per request, model, team, and feature. Set budgets, get alerts at 80%, and let Explane automatically route to cheaper models when you're approaching limits.

  • Per-request cost breakdown: prompt tokens, completion tokens, markup
  • Tag requests by team, product feature, or user segment for chargebacks
  • Projected monthly spend with confidence intervals
Spend by provider June 2025 · $1,840 total
$124
OpenAI
$74
Anthropic
$49
Gemini
$29
Mistral
$19
Bedrock
OpenAI · gpt-4o + gpt-4o-mini 67% $1,232
Anthropic · claude-3.5-sonnet 22% $405
Others · Gemini, Mistral, Bedrock 11% $203
Live Operations

A real-time view of
every request in flight.

The operations wall shows your AI traffic as it happens — model calls, token counts, latencies, error codes. Filter by provider, team, or status in under a second.

  • Sub-second latency on the live feed — no polling, pure WebSocket
  • One-click drill-down to the full trace for any log row
  • Search and filter by model, status code, team, or custom metadata
explane · live feed
streaming · press any key to pause
Intelligent Alerting

Catch regressions
before your users do.

Explane watches latency, error rates, cost anomalies, and provider health continuously. Smart alerts fire only when they matter — no alert fatigue, just signal.

  • Threshold, anomaly-detection, and budget-based alert types
  • Route alerts to Slack, PagerDuty, email, or webhook
  • Automatic suppression for known maintenance windows
Alert Feed 2 ACTIVE
Latency spike · gpt-4o
avg 2.4s · threshold 1.0s · 18 requests affected
2 minutes ago
CRIT
Budget 84% consumed · Engineering
$840 of $1,000 · at current rate, limit in ~6 days
14 minutes ago
WARN
Rate limit approaching · OpenAI
tier-2 TPM at 78% · auto-scaling to tier-3
1 hour ago
WARN
Failover resolved · Azure GPT-4
azure-gpt-4 503 → gemini-2.0-flash · recovered in 4s
2 hours ago
OK
Cost anomaly resolved
embedding cost spike traced to batch job · closed
6 hours ago
OK

Start seeing everything.

Full observability in under 10 minutes. No agents, no SDKs required — just point your requests through Explane.

Talk to Sales