Observability that can prove what it reports.
Most dashboards impute — they average over the gaps and hand you a confident number. Tranquility watches your agents, tools, and APIs across 173 frameworks, keeps traces, metrics, and logs in one lens, and turns every headline figure into a signed, offline-verifiable artifact that states how complete the window behind it was. Nothing is imputed; a health tag is calibrated, not a bare score.
Signals, agents, operations, evidence, economics — and cognition.
One evidence core underneath all of them, so a span, an SLO, a cost figure, and a guardrail reading are all the same kind of checkable record.
Signals
Traces, metrics, logs — one lens.
OTLP-native ingestion of spans, metrics, and logs side by side, keyset-paginated and correlated by trace. Every window carries its sampling coverage, so a number is never quietly extrapolated from a partial view.
Agents
173 frameworks, one normalized stream.
Invocations, tool calls, memory writes, sub-agent spawns, and channel messages — normalized across 173 agent frameworks. Run trees as DAGs, per-tool reliability and latency, and clustered failures, derived from what was actually observed.
Operations
Any-denomination SLOs.
One SLO engine for latency, error rate, cost per outcome, or mandate burn — with budgets, burn windows, and alerts. Every firing is sealed as evidence, so an incident's timeline can be proven after the fact, not just narrated.
Evidence
Every headline number, signed.
Usage and coverage seals turn each figure into a signed, offline-verifiable artifact that states both the number and how complete the window behind it was. "99.98% over the last 30 days" becomes something a customer can check, not take on faith.
Economics
Cost per resolved outcome.
The Goal Ledger decomposes spend into the outcomes it resolved — and the waste it didn't — with the decomposition shown, never guessed. You see what an agent's autonomy actually costs per result.
Cognition
Is the agent still on-charter?
Guardrail pressure mapped clause by clause, pre-execution blast radius, observed-vs-charter conformance, and vocabulary drift resolved against your Driftvice canon — the behavioral signals a span count can't show.
The number is only worth as much as its honesty about the gaps.
Nothing imputed
Success rates over known-outcome calls only; p50/p95 over timed calls only; a timed/untimed coverage split shown beside every figure. No silent averaging over gaps.
Calibrated, not scored
An agent's health is a tag — healthy / degraded / quiet — carrying a confidence and its basis, never a bare number that hides how little was observed.
Signed, verifiable evidence
SLO evaluations, usage figures, and coverage statements are Ed25519 artifacts. Anyone can verify them offline against the published key — observability you can check.
Resolved against your canon
Vocabulary drift and charter conformance are judged against your own governed definitions (Driftvice), not a generic model's guess at what your terms mean.
Point your agents' OTLP exporter at Tranquility.
Traces, metrics, and logs in one lens — with every headline number signed and verifiable. It's the same observability that backs our own public status page.
Trustvice 360Dichiara le regole. Fai controllare ogni esecuzione.
Agenti, prove, governance, commercio e informativa sono in Trustvice 360. Registrati gratis e apri tutte le suite da un'unica dashboard.