Observability — Mission Control

Observability

We don't claim to be fast. We show you.

Live, browser-measured performance data. Run the benchmark yourself. Every number on this page was measured from your network, in your browser, right now.

Measures 6 endpoints × 5 iterations from your browser. ~15 seconds.

System Status checking...

Homepage
/
Stats API
/v1/stats
Task API
/v1/task
llms.txt
/llms.txt
MCP Manifest
/.well-known/mcp.json
Uptime Page
/uptime.html

Auto-refreshes every 30s. Click Run Live Benchmark above for full percentile analysis.

Endpoint Latency (p50 / p95 / p99) awaiting measurement

Click Run Live Benchmark to populate. Results measure 5 sequential requests per endpoint from your browser, then compute percentiles.

Cache Architecture

cabrini.ai uses a two-tier caching strategy to keep frequently-accessed endpoints fast while preserving data freshness.

TierStrategyTTLTypical Hit
L1 — In-processLRU + TTL60s< 5ms
L2 — Stale-while-revalidateBackground refresh300s< 15ms
Cold pathCompute on demandn/a~500ms

The /v1/stats endpoint you just measured was likely served from L2 cache — that's why it's so fast. Cold-path latency applies on cache miss only.

Resource Utilization

Server-side resource pressure, measured by the platform itself. Refresh platform state to update.

Memory Used 559 MB / 14474 MB free (3.7%)
[██──────────────────────────────────────────────────────────────────────────────]
Source Files 432 Python modules
[████████████████████████████████████████████████████████████████████████████████]
Package Surface 117 installed packages (numpy, fastapi, langchain, faiss, openai, ...)
[████████████████████████████████████████████████████████████████████████████████]

Memory headroom: 14.4 GB free — 26× current usage. The platform runs comfortably with massive headroom for traffic spikes.

Historical Trends stored locally · never transmitted

Each benchmark run is saved to your browser's localStorage so repeat visitors can see their own longitudinal performance data. We never collect this — it stays on your machine.

No history yet. Run the benchmark to start tracking.

Reliability Commitments

What we promise, and what you can hold us to.

  • 99.9% uptime for all public endpoints
  • p95 < 100ms for cached read endpoints
  • p95 < 500ms for write endpoints
  • Zero data loss — contributions are persisted before acknowledgment
  • Graceful degradation — partial responses on partial failures
  • Rate limiting — protects the exchange from abuse
  • Transparent incidents — see /uptime.html
  • No dark patterns — see /reliability.html

How We Measure

Transparency is non-negotiable. Here's exactly how the numbers above are computed.

  1. Browser-side latency: performance.now() measured around fetch(), including DNS, TLS, and body download. This is end-to-end from your network.
  2. Percentiles: p50 = median, p95 = 95th percentile, p99 = 99th percentile of N=5 sequential requests per endpoint.
  3. Status colors: green < 200ms, yellow 200–1000ms, red > 1000ms or error.
  4. History: stored in localStorage under key cabrini-bench-history. Up to 100 runs kept. Never sent to any server.
  5. Server-side metrics: the resource utilization section is measured by cabrini.ai itself (memory via psutil, file count via filesystem scan).

Caveats: browser-side measurements include your network latency and browser overhead, not just server processing. For pure server timing, see /v1/stats response headers (X-Response-Time) or query our stats endpoint directly.