Sailhouse

Topics are the buckets your events are grouped into. You publish a JSON event to a single topic, identified by a slug, and the platform fans it out to every subscription attached to that topic. Events carry metadata too, which is how the platform attaches trace IDs or tags a cron-triggered event with its schedule. Topics and subscriptions are the two core resources, and everything else builds on them.

Sailhouse supports both delivery models. A push subscription sends a POST request to your endpoint when an event arrives, and expects a 200 response within 5 seconds. A pull subscription flips it: your worker asks for events on its own schedule, which suits processing-heavy jobs like a billing cycle or a weekly summary. Teams pick per subscription, so a single topic can feed a real-time webhook and a batch worker at the same time.

Wait groups are the feature that makes parallel agent work practical. You hand the platform a list of events to fire and a topic to notify on completion; Sailhouse runs them and publishes one event when the whole group is done. Without this, you'd write duplicate checks or a race-prone "is everything finished yet" loop. It's the same pattern as a wait-group in a programming language, moved out to an event-driven system where retries don't cause extra calls.

Sailhouse creates resources on demand. Send an event to a topic that doesn't exist and the platform makes it using your slug, then stores the event against it. Pull subscriptions can be created the same way, with a time window of up to 24 hours that pulls in everything from a topic over that period. The goal is fewer schema steps between writing code and shipping, which matters when you're iterating on an agent.

You can queue work for later in two ways. Scheduled events set a time on a single event or a subscription, handy for processing a payment 14 days after an order or sending a reminder. Cron jobs, currently in private beta, fire events into a topic on a repeating schedule, all in UTC, for daily roundups or hourly usage calculations. Both reuse the same subscription model, so there's no separate scheduler to learn.

Subscription filters let you keep topics generic while cutting noise and cost. Instead of splitting a topic into a dozen narrow ones, you filter at the subscription with a JSON path, so a consumer only sees the events it cares about. The platform also supports rate limiting and deduplication in the same layer, which is where you'd put guards against duplicate deliveries or a burst from a misbehaving publisher.

Sailhouse ships first-party SDKs for TypeScript and JavaScript, Go, Python (beta) and Rust (beta), plus a CLI and a REST API for everything else. The observability side covers tracing events, so you can attach trace IDs and follow an event through the system. Because it's hosted, the ingest regions currently span London, Ashburn and San Jose, with data written centrally and encrypted in transit and at rest.