OpenBridge is a macOS-native, local-first AI agent chat application designed as a privacy-focused alternative to cloud-based coding and productivity assistants like Claude Cowork and Codex. It empowers users to run AI agent sessions directly on their machine, leveraging their own API keys and models for enhanced security and control. The application targets developers and power users seeking a transparent, extensible, and safe environment for AI-assisted tasks.

How It Works

OpenBridge employs a hybrid architecture combining a native SwiftUI macOS shell with embedded React chat surfaces and a vendored kwwk agent runtime. It prioritizes local execution, storing credentials and agent state locally. A key innovation is its sandbox VM workflow, which isolates file operations within a virtualized environment. Users can review and approve all proposed file changes before they are applied to the host filesystem, mitigating risks associated with AI-driven modifications. The system manages agent sessions, provider authentication (supporting OAuth and API keys), model selection, and local skills, offering a self-contained agent experience.

Quick Start & Requirements

• Primary Install/Run: Requires cloning the repository, initializing submodules, building embedded web assets, building the macOS app via Xcode, and then launching the debug build.
  • git submodule update --init --recursive
  • cd web && yarn install --immutable && yarn build:embedded
  • cd ../sandbox-vm && make framework
  • cd ../macos && BUILD_CONFIGURATION=UnsignedDebug bash DevKit/Scripts/workspace_build_debug.sh
  • Launch: open -na "$(pwd)/.build/DerivedData/Build/Products/UnsignedDebug/OpenBridge.app"
• Prerequisites: macOS with Apple Virtualization.framework support, Xcode and command-line tools, Go 1.24+, Node.js with Corepack/Yarn. protoc is needed for VM RPC protobuf changes.
• Links: macOS app structure, embedded WebView, sandbox VM, architecture.

Highlighted Details

• Supports a wide array of AI providers including OpenAI, Anthropic, Google Gemini, Azure OpenAI, GitHub Copilot, and numerous others via direct API, OpenAI-compatible endpoints, or gateways.
• Features a macOS notch UI for tracking multiple agent tasks with live status indicators.
• "Computer Use" allows approved agents to operate local macOS applications.
• Extensible via local "Skills" packages, loaded from ~/.openbridge/skills or bundled sources.
• Sandbox VM provides per-session environments with overlay-backed file changes for review and acceptance/rejection.

Maintenance & Community

The README indicates the project is under active development, with current focus areas including the local agent loop, provider configuration, skill behavior, sandbox workflow, and native UI. No specific community links (Discord, Slack) or details on core contributors/sponsorships are provided.

Licensing & Compatibility

OpenBridge is released under the MIT License. This permissive license generally allows for commercial use, modification, and distribution, including within closed-source applications, with minimal restrictions beyond attribution.

Limitations & Caveats

The application is macOS-first and relies on Apple's Virtualization.framework, limiting its use to compatible macOS versions. It is explicitly stated to be under active development, suggesting potential for ongoing changes, bugs, or incomplete features as the core functionalities are refined.