vitel 0.1.0

Vitals for AI Agents — graded observability returning pass/warn/fail on metrics, SLOs and error budgets, with each finding grounded in a Metric. Sibling to AgentVision (eyes), Audel (ears) and Verel (brain).

vitel

Vitals / interoception for AI agents — graded observability that turns telemetry (metrics, SLOs, error budgets) into a pass / warn / fail verdict, with every finding grounded in a Metric (name, window, observed vs threshold, burn-rate).

vitel is a sense organ built on the shared agentsensory contract — sibling to AgentVision (eyes), Audel (ears) and Verel (brain). Drop it into CI/CD to gate a release on error-budget burn, or run it as a Verel sense so the brain won't mark a deploy "done" until post-rollout vitals confirm health.

What it does

Function	Behavior
check(source, *, slo=None)	deterministic threshold / SLO / burn-rate evaluation, no LLM — main path
analyze(source, *, slo=None, backend=None)	adds anomaly / LLM critique on top of check
watch(source, *, window, interval)	poll / stream over time; trend, degradation, liveness
render(source)	normalized series + rates / percentiles / burn-rate (the trustworthy signal)

Install

pip install vitel                      # light base: JSON/CSV series + thresholds
pip install "vitel[prometheus]"        # PromQL / Prometheus
pip install "vitel[otel]"              # OpenTelemetry (OTLP)
pip install "vitel[cloud]"             # Datadog / CloudWatch
pip install "vitel[psutil]"            # process self-vitals
pip install "vitel[all]"               # everything

License

MIT © Amit Patole