Provia

The network that assembles itself

Provia Fabric is built on the self-organizing Tessera routing protocol, and it assembles your application's cluster network autonomously, node by node, region by region. A node boots, proves its identity, and announces itself. From that moment traffic can find it, with zero registries to update and zero load balancers to babysit.

get early access see the platform
tessera

Powered by Tessera

Provia Fabric runs on Tessera, the routing protocol built for exactly this job. It travels with every node Provia creates, so the moment a machine exists, the network already knows how to reach it. Nothing to stand up. Nothing to register. Nothing to babysit at three in the morning.

  • Every node arrives already connected, every time
  • Traffic finds the closest healthy capacity on its own
  • Your app never has to know where anything lives
  • It keeps running whether or not anyone is watching
tessera, on every nodea node appears already reachable a node dies traffic already moved a region opens findable from everywhere # nothing to wire up. not once, not ever.
self-registration

Announcing is the registration

There is one source of truth for what should exist, and it lives in Provia. Everything live happens in the fabric itself: a node advertises a signed route about itself, peers verify it against its identity, and the route spreads by gossip.

  • New capacity is findable the moment it finishes booting
  • Identity and authorization ride in certificates, checked on every advertisement
  • Unknown nodes are refused their bootstrap: the roster fails closed
registration = advertisementn03 boots -> advertises route(self), signed peers verify claims -> accept -> gossip n03 receiving traffic t+seconds
liveness

Dead nodes remove themselves

Every route carries a short lease its owner keeps refreshing. A node that stops refreshing falls out of the routing table in seconds, and its traffic re-settles onto healthy peers. Registration and deregistration are the same mechanism running forward and backward.

  • Peer-to-peer failure detection in under a second
  • Graceful shutdown drains first: capacity leaves only after traffic has moved
  • Health scripts, cron probes, and cleanup jobs all become unnecessary
a node dies at 14:02:0714:02:07 n02 stops refreshing its lease 14:02:08 peers detect the silence 14:02:10 routes re-settled on n01, n03 14:02:11 traffic flowing, pager quiet
autonomy

Routing survives everything

The control plane hands a node its peer set and policy exactly once, at boot, and the node caches it. After that the fabric runs on its own: forwarding, failover, and placement all happen between the nodes themselves.

  • A control-plane outage pauses new bootstraps and policy edits. Live traffic keeps flowing
  • Every node holds the map and reroutes locally, with zero round trips
  • Policy changes roll out on the nodes' own refresh cycle
dependency budgethot path nodes only boot path control plane (once, then cached) policy path control plane (pulled, ~2 min) # the fabric needs the control plane for # nothing once a node is running
postures

Consulted once or on every request

Each service picks how deeply the fabric participates. Long-lived media sessions get placed once and then flow peer to peer. Stateless APIs get routed on every request. Streams that must survive failover flow direct while the fabric stands ready to re-place them.

  • Direct: place the session, then step out of the way
  • Gateway: route every request, enforce policy per call
  • Assisted: direct-flow speed with mid-session re-placement
per-service choicemedia direct # place once, flow p2p api gateway # every request routed broadcast assisted # direct + re-placeable
compliance

Residency fails closed

Declare that a service's data stays in a jurisdiction and the fabric enforces it in the routing layer itself. If every compliant backend disappears, the fabric withdraws the route rather than spilling traffic somewhere it must never go.

  • Residency is enforced mid-failover, exactly when other systems leak
  • Policy is declared once and every node pulls it
  • Placement decisions honor jurisdiction before cost, every time
the eu backends all just diedserve(X, residency=EU)? eu-1 gone · eu-2 gone us-1 healthy but outside the fence decision: withdraw the route, alert # degraded beats non-compliant, always
edge

One front door for the world

The global edge tier puts an anycast front door ahead of your fleet: users land on the nearest point of presence and ride the fabric to the closest healthy backend. Shipped clients keep a single stable hostname through every scale event, failover, and migration underneath.

  • Geo and residency routing at the edge
  • Raw protocol support beyond HTTP, including UDP workloads
  • Available as an add-on for Growth and Enterprise
what ships in your appapi_base = "acme.api.provia.app" # that string outlives every node, region, # failover, and cloud underneath it

Provia creates the nodes.
The fabric makes them findable.

Intent lives in the control plane. Liveness lives in the network. Each layer does the one job it is built for, and your traffic never waits on either.

get early access see pricing