Introduction

SACP (Symposium's extensions to ACP) is an SDK for building composable AI agent systems using the Agent-Client Protocol.

What is SACP?

SACP extends the Agent-Client Protocol (ACP) to enable composable agent architectures through proxy chains. Instead of building monolithic AI tools, SACP allows you to create modular components that can intercept and transform messages flowing between editors and agents.

Think of it like middleware for AI agents: you can add logging, inject context, provide additional tools, or modify behavior - all without changing the editor or base agent.

Key Capabilities

Proxy Chains: Chain together multiple components, each adding specific capabilities
Message Transformation: Intercept and modify requests and responses
Tool Injection: Add MCP servers and tools to any agent
Editor Agnostic: Works with any ACP-compatible editor (Zed, Claude Code, etc.)
Agent Agnostic: Works with any ACP-compatible agent

Example Use Case: Sparkle Integration

Consider integrating Sparkle (a collaborative AI framework) into a coding session. Sparkle needs initialization and provides MCP tools.

Without SACP: Manual initialization each session, or agent-specific modifications.

With SACP: A Sparkle proxy component automatically:

Injects Sparkle's MCP server during initialization
Prepends the embodiment sequence to the first prompt
Passes everything else through transparently

The editor sees a normal ACP agent. The base agent has Sparkle tools available. No code changes needed on either side.

Architecture Overview

flowchart LR
    Editor[ACP Editor]
    
    subgraph Conductor[Conductor Process]
        P1[Proxy 1]
        P2[Proxy 2]
        Agent[Base Agent]
        
        P1 -->|messages| P2
        P2 -->|messages| Agent
    end
    
    Editor <-->|ACP| Conductor

SACP introduces three types of components:

Conductor: Orchestrates the proxy chain, appears as a normal ACP agent to editors
Proxies: Intercept and transform messages, built using the sacp-proxy framework
Agents: Provide base AI model behavior using standard ACP

The conductor manages message routing, making the proxy chain transparent to editors.

Who Should Use SACP?

Proxy Developers: Build reusable components that add capabilities to any agent
Agent Developers: Create specialized agents that work with any ACP editor
Client Developers: Build ACP-compatible editors and tools
Integration Developers: Connect AI agents with existing systems and workflows

Repository Structure

This repository provides three core crates:

sacp: Core protocol types and traits for building clients and agents
sacp-proxy: Framework for building proxy components
sacp-conductor: Binary that orchestrates proxy chains

Getting Started

Read the Architecture Overview to understand how SACP works
See the Protocol Reference for technical details
Follow Building a Proxy to create your first component
Check out the SACP RFD for the complete specification

Relationship to ACP

SACP is an extension to ACP, not a fork. SACP components communicate using ACP's extension protocol (_meta fields and custom methods). Standard ACP editors and agents work with SACP without modification - they simply see a normal ACP agent when talking to the conductor.

SACP: Symposium's Extensions to ACP