Synadia Agents Protocol for NATS Released

What changed

Synadia has introduced the Synadia Agent Protocol for NATS, a framework designed to enable AI agents to function as discoverable microservices within the NATS messaging system. This protocol facilitates communication between various AI models, including Codex, OpenCode, Claude Code, PI Agent, and others, allowing them to be prompted and stream responses using a consistent wire format.

The project provides official caller and host SDKs in both TypeScript and Python. The caller SDKs (@synadia-ai/agents for TypeScript and synadia-ai-agents for Python) enable applications to discover and interact with registered agents. Conversely, the host SDKs (@synadia-ai/agent-service for TypeScript and synadia-ai-agent-service for Python) allow developers to build and register their own agents on NATS.

In addition to the SDKs, the repository includes pre-built channel plugins that integrate popular AI harnesses directly onto NATS without requiring custom code. These plugins register as agents microservices and expose prompt, status, and heartbeat endpoints. Supported agents include Claude Code (cc), OpenClaw (oc), PI Agent (pi), Hermes, DeerFlow (df), Flue, Eve, open-agent, OpenCode (opencode), Codex (codex), and ACP-based agents.

Examples are provided to demonstrate end-to-end applications, such as a Vue 3 browser client for agent discovery and prompting, headless session management for Claude Code and PI agents, and custom agents built from scratch using DSPy for ReAct and deep-research capabilities. These examples showcase features like streaming responses, mid-stream queries, and tool usage.

The wire protocol defines a clear structure for requests (plain text or JSON with optional attachments) and streamed responses (typed JSON chunks for content, status, and mid-stream queries), with an empty-body terminator for completion. Errors are communicated via NATS error codes. Agent discovery leverages standard NATS microservice conventions.

Why it matters for builders

This initiative provides a robust framework for building distributed AI systems. Developers can now integrate a wide array of AI models into their NATS-based applications, treating them as interchangeable services. The availability of both caller and host SDKs, along with pre-built agent plugins, significantly lowers the barrier to entry for creating sophisticated AI-powered workflows.

Builders can experiment with different AI models by simply plugging them into the NATS network via the provided channels, or they can develop custom agents tailored to specific needs. The protocol's emphasis on standardization ensures interoperability, allowing for seamless communication and data exchange between diverse AI components.

Practical impact

Developers can begin integrating AI agents into their NATS applications immediately by installing the respective SDKs. For instance, a TypeScript developer can use @synadia-ai/agents to discover and prompt an agent, while a Python developer can use synadia-ai-agents for similar functionality. Those looking to expose their own AI models as services can utilize the host SDKs to register custom agents.

Experimentation with the provided examples, such as the agent-web-ui or headless session managers, can offer practical insights into building complex agent interactions. The project's latest release, python-v0.7.1, indicates ongoing development and refinement.

Caveats and source limits

The provided source is a GitHub repository README, offering detailed technical information but lacking independent benchmarks or performance metrics. Specific pricing for any associated Synadia services is not mentioned. While the repository indicates a "fresh release," a precise release date is not explicitly stated beyond the published_at timestamp in the metadata, which appears to be a future date. The repository has 50 stars and 2 forks, suggesting early community adoption.