Agent Communication Protocol

Agent Communication Protocol (ACP) is an open standard designed to enable seamless communication between AI agents regardless of framework, programming language, or runtime environment. ACP is lightweight, HTTP-native, and requires only a minimal setup for agents to send and receive messages to each other with a consistent interface.

ACP was developed in response to the increased fragmentation of a rapidly growing agent ecosystem. Modern AI systems often rely on modular agents which perform tasks such as retrieval, reasoning, classification, or tool use. Yet these agents are typically locked into isolated frameworks, making it difficult to discover, reuse, or integrate them across platforms. This fragmentation leads to duplicated work, an inconsistent developer experience, and one-off integrations that don’t scale. ACP addresses these challenges by standardizing how agents communicate while letting developers decide how they’re built. This allows agents developed in different environments interoperate through one unified RESTful interface.

Use Cases

• Dynamic Updating: By making agents ACP-compatible, organizations can easily replace or upgrade agents, even if they are built with different technologies. This allows systems to stay flexible and interoperable, making it easier to test and deploy the agents that deliver the best results.
• Specialized Agents can work as a Team: Instead of building one large, all-purpose agent, ACP enables smaller, specialized agents to collaborate in real time. For example, a research agent, a visualization agent, and a financial modeling agent can work together to generate a report. Each agent handles its part of the task and passes it along, just like members of a human team working together toward a shared goal.
• Cross- Company Workflows: Most organizations rely on a wide range of systems, including customer support platforms, inventory management tools, HR systems, and more. With ACP, each of these systems can be supported by a specialized agent. For instance, if a customer submits a complaint about a delayed shipment, the customer support agent can request assistance from the inventory agent. Each agent focuses on its own domain but communicates clearly with others through ACP, enabling smooth workflows across the entire organization.
• Inter-Organizational Collaboration: ACP also opens the door to secure collaboration between agents from different companies. This makes it possible to move beyond internal use cases and build agent-based workflows that span organizations. By allowing agents from separate systems and companies to work together, ACP helps create entirely new models for partnership and shared innovation.

Core Features of ACP

• REST-Based Design: ACP is built on REST principles, exposing well-defined HTTP endpoints for sending, receiving, and routing agent messages. This ensures compatibility with common web infrastructure and minimizes the need for specialized tooling.
• SDK-Optional Architecture: ACP can be used directly via HTTP clients like curl or Postman, lowering the barrier to entry for integration. An optional SDK is available to streamline development, but not required
• Multimodal Message Support: Messages in ACP can contain structured data, plain text, images, or embeddings, making the protocol suitable for complex workflows involving LLMs, vision models, or hybrid systems.
• Asynchronous by Default, Synchronous Supported: ACP is designed to handle asynchronous communication out of the box, which is ideal for long-running or multi-step tasks. But it also supports synchronous messaging for low-latency use cases, like interactive applications. ACP handles streaming using Server-Sent Events (SSE) over HTTP.
• Offline and Secure Discovery: ACP supports metadata-based discovery of agents, even when they’re not currently running or connected to the network.

Governance and BeeAI Ecosystem

ACP is developed under the Linux Foundation to ensure open governance and transparency. Its primary implementation is provided by the BeeAI Framework, a production-grade framework for building ACP-compliant AI agents in either Python or TypeScript. These agents can then be discovered, run, and shared on the BeeAI Platform, which provides the infrastructure layer for production deployments.