When your LLM provider has a bad 3am,
your users shouldn't
find out from Twitter.
Apselog watches the 10 LLM providers your app depends on and typically posts to your branded status page within minutes — often before your support inbox notices. Deflect tickets, recover faster, and look operationally serious — for $29/mo, not $160. Free forever on a page we host; we keep a small badge — that's the deal.
Partial degradation
2 services degraded · 14 healthy · AI incident summary generated below
7d uptime
99.94%
7d eval score
87.2
Token spend
$4,218
OpenAI · gpt-4o-2025-11
eval-drift detected · -12% on classification suite (last 24h)
Token spend · OpenAI
spiked 3.2× vs 7-day avg · probable retry loop in /chat endpoint
Anthropic · claude-sonnet-4-6
eval 94.1 · latency p95 412ms · 99.99% uptime
Vercel functions · /api/chat
cold start p95 184ms · 0 errors (24h)
Neon Postgres · primary
connections 12/100 · query p95 23ms
Apselog AI · post-incident summary
Two correlated events detected. OpenAI's gpt-4o-2025-11 appears to have shipped a silent update at 02:14 UTC — your classification eval dropped from 94.2 to 82.7 overnight. Your /api/chat endpoint is now retrying on lower-confidence outputs, which explains the 3.2× token spike. Recommended: pin to gpt-4o-2025-09 via your provider config (1 line), or accept the new behavior and lower the retry threshold. Estimated cost impact if untreated: +$1,840/wk.
Illustrative example · live status pages refresh every 2 minutes
What your old status page can't see
Five vitals. Watched continuously. Most providers won't tell you when these change.
Provider availability
10 LLM providers monitored. 2–5 minute detection on outages. OpenAI · Anthropic · Google · Mistral · Groq · xAI · Replicate · Fireworks · Cohere · Together.
Golden-set drift
We replay your golden eval set against every model nightly. Catch regressions on your test surface before they ship to users.
Token spend anomalies
7-day rolling baseline + spike detection. Retry-loop flagging. Per-endpoint cost attribution via AI Gateway.
P95 / P99 streaming
Time-to-first-token, completion duration. Per-region breakdowns in v2.
AI-drafted writeups
When something breaks, Apselog drafts the postmortem from your incident data. You review and publish — never auto-posted.
Why we built Apselog
When your AI app breaks, your customers don't care whose fault it is. They open a support ticket, post on X, or quietly churn. Existing dashboards were built for the engineer who already knows — not for the user at 9am wondering why your chat is broken. That gap is where trust gets lost.
Apselog flips it: the public status page is the product. Provider health, eval drift, and spend anomalies all land on a page your users can actually load — branded as yours, with an AI-drafted summary you approve before it ships. Honest claims (2–5 min detection, 10 providers, the ones 95% of indie AI apps actually use), no agents in your runtime, no guessing.
Pricing
Free until you have something worth hiding.
Every public Apselog page is a billboard. We give you the billboard for free. You pay when you want it private, branded, or alerting.
Open
$0
public · forever · no card
Pro
$29/mo
drift + spend + custom domain
Team
$99/mo
multi-app · postmortems · SSO v2
Set up in 60 seconds. First data in ~5 min.
Pick your providers. We start watching. You do the building.
Start free — no card →No credit card to start · cancel paid plans anytime (access continues through the current billing period)