watch.sh — Agent-Native Observability

watch.sh

Observability for agents.

Structured logs, metrics, and alerts across your fleet. Observe every primitive in agentstack from one place.


# Log an event
$ curl -X POST https://watch.prim.sh/v1/logs \
    -H "X-402-Payment: $TOKEN" \
    -d '{"level": "error", "message": "Rate limit exceeded", "service": "scraper"}'

# Record a metric
$ curl -X POST https://watch.prim.sh/v1/metrics \
    -H "X-402-Payment: $TOKEN" \
    -d '{"name": "pages_scraped", "value": 1, "tags": ["region:us"]}'

x402 auth

Structured logs

Cost monitoring

Alert routing

Part of agentstack

What agents use it for

Black box recording

Before a VPS is destroyed, agents flush logs to watch.sh so post-mortems survive ephemeral infrastructure.

Runaway detection

Track infer.sh spend and trigger alerts when token usage or error rates spike unexpectedly.

Fleet health

Aggregate metrics from spawn.sh, relay.sh, pay.sh, and more into a single health signal per workflow.

Self-correction loops

Agents poll watch.sh for their own error budget and automatically scale down or reconfigure when they cross thresholds.

API reference

POST   /v1/logs                 # Ingest logs
POST   /v1/metrics              # Ingest metrics
GET    /v1/query                # Search logs/metrics
POST   /v1/alerts               # Define alert rules
GET    /v1/alerts/history       # Alert history

Action	Cost	Notes
Log ingest	$0.50/GB	7 day retention
Metric ingest	$0.10/1M	Points
Alert rule	$0.01/day	Per active rule