Use this command to install agentbridge with WinGet:
winget install --id=AGNTCY.agentbridge -e
AgentBridge is a CLI binary designed to facilitate secure communication and collaboration between AI agents. Built as a general-purpose agent interconnect, it enables seamless interaction across different environments while prioritizing security and efficiency.
Key Features:
Secure Communication: Employs the SLIM protocol for authenticated and encrypted messaging between agents.
Sandbox Integration: Leverages SHADI's sandboxing capabilities to ensure secure execution and isolation of processes.
Cross-Platform Compatibility: Works across Linux, macOS, and Windows, with streamlined deployment via Winget.
Agent Interoperability: Supports integration with popular AI tools like Copilot, Claude Code, and Codex.
CLI Accessibility: Provides a command-line interface for easy scripting and automation of agent workflows.
Target Audience and Benefits:
Ideal for developers, DevOps engineers, and security teams looking to enhance collaboration between AI agents. AgentBridge enables organizations to maintain secure, efficient communication while minimizing risks associated with traditional inter-agent interactions.
README
SHADI
Secure Host for Agentic AI Dynamic Instantiation (SHADI) is a secure runtime for AI agents.
It gives agents a safer place to run by combining identity verification, gated secret access, OS-level sandboxing, encrypted local memory, and secure messaging.
Why SHADI
SHADI is for teams that want agents to work with real tools and real credentials without treating the host machine as a fully trusted environment.
Verify who an agent is before releasing secrets.
Constrain what a process can read, write, execute, and reach on the network.
Keep local memory encrypted at rest.
Connect agents over authenticated SLIM messaging.
Audit changes with snapshots and runtime inspection tools.
What You Get
shadictl: the main CLI for policy, sandbox execution, identity, secrets, memory, and shell control.
agentbridge: a general-purpose agent interconnect over A2A — coding CLIs (Claude Code, Copilot, Codex, Cursor Agent) are the flagship use case.
Each Homebrew formula builds from its own published release tag
(agntcy-shadi-cli for shadictl, agntcy-agentbridge-cli for agentbridge).
Both releases also include prebuilt archives for Linux (x86_64 and
aarch64), macOS (arm64 and x86_64), and Windows (x86_64).
For unreleased changes or any other host, use the source build flow below.
Quick Start
cargo build --workspace
cargo test --workspace
cargo run -p shadi_demo_bot -- feature-bot
The demo bot runs a compact end-to-end check across secrets, memory, sandboxing, and local SLIM messaging.
