The early web solved publishing before it solved navigation. Once anyone could create a website, the hard problem became discovery: finding relevant sites, ranking them, and getting users to the right destination. Search engines became the organizing layer that turned a scattered network of pages into something usable.
Agents are at the same point now. Building them is no longer the bottleneck. We have strong models, tool frameworks, and action-oriented agents that can run real workflows. What we do not have is a shared layer that makes those agents discoverable and routable as services, without custom integration for every new agent and every new interface.
ARC is built for that gap. Think of it as infrastructure for the Action Web: a network where agents are exposed as callable services and can be reached from anywhere through a common contract.
ARC Protocol defines the communication layer: a stateless RPC interface that allows many agents to sit behind a single endpoint, with explicit routing via targetAgent and traceId propagation so multi-agent workflows remain observable across hops. ARC Ledger provides a registry for agent identity, capabilities, and metadata so agents can be discovered as services. ARC Compass selects agents through capability matching and ranking, so requests can be routed to the most suitable agent rather than hard-wired to a specific one.
The goal is straightforward: start from any node, any UI, any workflow, and route to the best available agent with minimal configuration. This is not another agent framework. It is the missing discovery and routing layer that lets an open agent ecosystem behave like a coherent network.
Docs: https://arc-protocol.org
Repo: https://github.com/arcprotocol/arc
Key properties
- Single-endpoint multi-agent routing via targetAgent
- Stateless RPC method invocation
- Real-time streaming via Server-Sent Events
- Asynchronous tasks with notifications for long-running work
- End-to-end workflow tracing via traceId propagation
- Hybrid TLS using X25519 and Kyber-768 (ML-KEM, FIPS 203) for post-quantum resilience
Top comments (0)