See, replay, prove

Observability is
the audit log.

Every brane emits structured spans, signed metrics, and reproducible compute artifacts — VM snapshots, framebuffer captures, full audit chains. The stack is OpenTelemetry-native; the backends are yours. Nothing the substrate sees, you can't see — and the signature chain proves it.

[telemetry]
OpenTelemetry · OTLP
[storage]
Tempo · Loki · Prometheus · Grafana
[artifacts]
snapshot.vm · screenshot.fb · trace.span
[chain]
hash-chained · ed25519-signed
At a glance

The properties
that matter.

100%
spans signed

Every span carries the brane Ed25519 signature plus the substrate co-signature.

120 ns
span overhead

Inline span emission costs ~120 nanoseconds per event. No sampling required for most workloads.

OTLP
native protocol

Speaks OTLP grpc and OTLP/http natively. Drop in any OpenTelemetry collector.

replays from artifacts

Snapshot + framebuffer + trace = a session you can replay forever.

In the manifest

How you
describe it.

observability.zgraph.toml
[brane.api]
substrate   = "gaia"
capability  = "standard"
image       = "omega/api:1.4"
// telemetry pipeline + artifact policy

// telemetry pipeline + artifact policy

In operation

What it
looks like running.

$ω trace https://api.branes.sh/score?id=42
→ trace.id b3:9af2c8... api.branes.sh 4ms gateway · cloudflare-front gaia-04.brane.api 42ms cold=1.2ms · cap=net.http └─ score-edge 18ms queue=8 → score score.0017 92ms embed + rerank …
$ω artifacts replay snap-20260501-1341 --substrate gaia --capability frontier
→ resumed VM in 121ms · framebuffer attached · trace re-attached → run identical to original within 2µs of wall time
protocol
OpenTelemetry · OTLP/grpc + OTLP/http
signing
every span signed; chain-of-custody preserved
sampling
configurable; default 1.0 (no sampling)
metrics
Prometheus-compatible; OTLP-native
logs
structured (key/value), Loki-ready
artifacts
snapshot.vm · screenshot.fb · trace.span
By design

What this
surface does.

Every event signed twice

Brane key + substrate key. You can prove a span came from your code on our hardware (or yours), and that the substrate received it intact.
PROVENANCE

Reproduce a session

Compute artifacts (snapshot + framebuffer + trace) can be re-attached to a fresh substrate. The session resumes inside 200ms; the trace continues from where it stopped.
REPLAY

Bring your own backends

Tempo, Loki, Prometheus, Grafana, Honeycomb, Datadog — all speak OTLP. The substrate is the producer; you own the storage.
OPEN

Hash-chained operational log

Spawn, kill, capability change, scale event — every lifecycle action enters a hash-chained audit log. Anchor it to Sigil for external verification.
AUDIT