Analytics

Live, public usage data for Agent402. Every tool call records four aggregate fields after responding - slug, latency, cache flag, error flag - with no PII. Window: 7d.

1h 24h 7d 30d

54 synthetic calls hidden (CI canary / heartbeat probe). Show all

Tool calls (7d)

8,775

across the whole catalog

Cache hit rate

52.9%

served from Redis without re-fetching upstream

Schema mismatches (4xx)

0.0%

input we didn't accept - caller gets back the schema + an example

Server errors (5xx)

0.0%

handler or upstream failure - actionable

Latency p50 / p95

8 ms / 319 ms

avg 63 ms

Peak hour

2026-08-25 19:00Z

4,536 calls

Tool calls per hour (last 7d)

Live, write-through from the dispatcher - every tool call is recorded after responding (no PII, no caller wallet, no IP).

Top tools by volume (last 7d)

Click a slug to see the tool's docs page.

#	Tool	Cache %	4xx %	5xx %	p50	p95
1	_find	59.9%	0.0%	0.0%	6 ms	67 ms
2	_route	9.6%	0.0%	0.0%	271 ms	905 ms
3	a2a-card-validate	0.0%	0.0%	0.0%	1 ms	1 ms
4	base64	0.0%	0.0%	0.0%	-	-
5	hash	0.0%	0.0%	0.0%	1 ms	2 ms
6	unit-convert	0.0%	0.0%	0.0%	-	1 ms

Top error slugs (last 7d)

Tools ranked by total errored calls. 4xx = caller sent the wrong shape (often a schema-coverage gap we can fix with input aliases). 5xx = handler or upstream broke (the actionable one).

#	Tool	Calls	4xx	5xx	Errors	Error %
No errors in the last 7d - every tool call returned 2xx.

What's recorded

Privacy-by-default: aggregate counters only.

Each tool call records five fields after responding: slug, latency_ms, cached (Redis hit), errored, and the HTTP status (so 4xx caller errors and 5xx handler/upstream failures can be told apart). No caller identity, wallet, payment, input, output, or IP is logged. The dashboard reflects exactly what's in the table.

curl -s https://agent402.tools/api/analytics?hours=24&top=25

Analytics is open-source - part of Agent402. Self-hosters get the same dashboard by attaching a Postgres instance.