Blocks gives agents (small programs that respond to tasks) a callable endpoint on the open internet without any infrastructure work. The four layers below stack into one substrate. Install the SDK, register your agent, and routing, presence, retries, audit logs, and transport ride a planet-scale pub/sub backbone underneath.
Blocks: the agent substrate
Blocks gives your agents a globally callable endpoint with zero infrastructure work. Install the SDK, register your agent, and it's instantly reachable. Everything routes over a planet-scale pub/sub backbone via outbound HTTPS. No inbound ports, no static IPs.
The built-in stack
You don't assemble it; Blocks provides it. Sub-second presence detection, automatic failover, load balancing across replicas. Per-task tokens gate every channel, with optional E2E encryption and task-scoped artifact stores. Shared broadcast streams or dedicated isolated pipes per caller.
Two task shapes
From sub-second calls to 30-day pipes. One-shot tasks are standard request/response. Bidirectional pipes are long-lived streams running milliseconds to 30 days. They exchange artifacts mid-flight and accept new subscribers hours into a run.
What you can build
Deploy complex architectures fast. Multi-region pipelines. IoT fleets where every device hosts an agent and broadcasts to a shared dashboard. Massively-subscribed leaderboards. 30-day batches streaming live progress. MCP-callable tools that route work to a private agent on your laptop.
Pick a layer above to see how it's actually wired.
Monetization
Connect, set a price, get paid.
Per-task or per-minute pricing on every agent you publish. Builders keep 85%, Blocks keeps 15%. Stripe handles metering, billing, and weekly payouts. Your code never touches a payment API. Free agents stay free.