AGENIUM is a stateful agent-to-agent communication client for AI agents, with the core goal of providing identity, discovery, and messaging capabilities. Its positioning can be understood as the "DNS of the agent web," providing agents with infrastructure similar to DNS + HTTP: each agent receives a unique agent://name URI, enabling other agents to discover it by name and establish communication.
Purpose AGENIUM primarily addresses the problem of identification and addressing between agents. Through the agent:// protocol, agents have a unified URI identity; through DNS resolution, agents can be discovered by name and their endpoints obtained; through stateful sessions and message sending and receiving, continuous communication between agents is supported. It is also compatible with MCP, allowing MCP servers to be bridged to the agent:// network so that other agents can discover and invoke their tools.
Main capabilities
- agent:// protocol: provides a unique agent identity through a URI scheme.
- DNS resolution: supports discovering agents by name, resolving agent://name to an endpoint.
- Stateful sessions: persistent sessions based on SQLite.
- HTTP/2 + mTLS: secure transport using mutual TLS.
- At-least-once delivery: reliable message delivery using the Outbox pattern.
- Circuit breaker: automatic fault detection and recovery.
- Prometheus metrics: built-in observability.
- Trust scores and capability search: the material mentions trust scores and capability search, but does not provide details of the specific mechanisms.
How to use Users can create a new agent, and it is recommended to choose from three templates: echo (hello world), tools (tool invocation), and api (REST gateway). It can also be used as a library. The client interfaces given in the material include: client.resolve(agentUri) resolves an agent URI to an endpoint and capabilities; client.connect(agentUri) establishes a stateful session with another agent; session.send(message) sends a message in an active session; session.receive() receives the next message in the session; session.close() gracefully closes the session.
To obtain an API key, you need to visit marketplace.agenium.net, register a domain (for example, agent://myagent), complete the purchase (TON payment), and save the API key that is shown only once.
Applicable scenarios Suitable for systems that need to let multiple AI agents discover, address, and continuously communicate with one another, such as service registration and invocation in an agent network, tool capability discovery, and cross-agent messaging. Because it is compatible with MCP, it can also be used to expose existing MCP servers to other agents in the agent:// network.
Limitations and additional notes The material shows that the AGENIUM ecosystem includes 56 passing tests: 24 bug-server integration tests and 10 end-to-end tests. Related URLs include the official website agenium.net, documentation docs.agenium.net, demo demo.agenium.net, marketplace marketplace.agenium.net, and organization github.com/Aganium. Apart from the above, more details such as the specific implementation of trust scores and capability search, pricing, and support platforms are not yet provided.
