The orchestration graph
Every event, routed with intent.
Aperture is the control plane for real-time data and AI-agent workflows. Connect sources, branch on logic, retry on failure, and watch the whole system reason about itself — as one live graph.
- events / sec
- 1,284,902
- p99 routing
- 17 ms
- active workflows
- 412
01 — What it does
A control plane, not another queue.
Connect
Fifty sources, one schema registry. Kafka, webhooks, change-data-capture, queues — validated and typed the moment they arrive.
Branch
Route on any predicate. Fork to an agent, fan out to N sinks, short-circuit on cost — logic lives in the graph, not buried in glue code.
Recover
Retries with backoff, dead-letter replay, and idempotency keys you never had to write. Failure is a state, not an incident.
Observe
Every packet traced end to end — latency, cost, and lineage per hop, queryable in plain SQL. The graph explains itself.
02 — Pipelines
Watch a packet cross the graph.
Sources fan into a router that branches on your logic — enrich, dedupe, retry, fork to an agent. Every hop is typed, observable, and replayable. Below is one tenant's live flow, not a diagram of one.
03 — Observability
Replayable by default.
Tail any tenant's graph live, or scrub history. Each line is a real event with its trace id, latency, and the decision that routed it.
- traced today
- 48.2M
- replays
- 1,204
12:04:01.221OK ingest kafka:orders → schema v8 validated
12:04:01.224OK branch cost-guard → under budget, continue
12:04:01.226WARNsink:agent timeout → retry 1 scheduled (120ms)
12:04:01.351OK retry 1 → 200, lineage written
12:04:01.353OK fan-out warehouse,alerting → committed
▍
04 — Speaks to everything you run
Start in minutes
Put your events on rails.
No credit card · SOC 2 Type II · cancel anytime