Signals with provenance
Metrics, checks, logs, events, quality, freshness, units, and source identity remain explicit. Missing data is not quietly interpreted as a healthy zero.
Open-source monitoring & operations
Know what changed.
Understand why.
Watchpost is a self-hosted monitoring platform designed around durable evidence, clear operational state, and an agent that can investigate your systems without inheriting unrestricted control of them.
In active development · architecture published early
Live signals
CPU42%
Memory68%
Disk /data86%
HTTP p95184ms
Evidence timeline
deploy api@8f21c7
DB pool saturation increased
latency rule firing
incident opened · evidence linked
Built for the whole incident
From observation to explanation.
Traditional monitoring often leaves the operator to connect dashboards, alerts, changes, logs, and runbooks by hand. Watchpost is being designed so those relationships are durable product objects rather than context scattered across browser tabs.
Deterministic rules
Duration, hysteresis, maintenance, dependencies, deduplication, acknowledgement, and resolution work without asking a language model what should be true.
An evidence-grounded agent
Ask what changed and why. The agent follows bounded operational evidence, shows its work, preserves uncertainty, and starts read-only.
Operational state, connected
A coherent path through failure.
Watchpost’s object model is intentionally small. Collection produces trustworthy observations; rules produce state; incidents preserve the investigation; actions remain separately authorised.
01Collect
Observe hosts, services, endpoints, applications, and events.
02Evaluate
Apply deterministic rules with explicit missing-data policy.
03Alert
Route and deduplicate state transitions without storms.
04Investigate
Join signals, changes, topology, logs, and history.
05Act
Use typed, scoped, approved operations with audit.
Agent-native, not agent-dependent
The model may explain an incident. It does not get to redefine authority.
Authentication, rule evaluation, retention, notification, approval, and audit remain ordinary deterministic software. AI operates inside those boundaries, treating monitored content as untrusted input and citing the evidence behind its conclusions.