EvoFlux is an open-source, local-first desktop workspace positioned as a collaborative environment for AI agent teams. It brings agent conversations, files, terminals, browsers, Git, and verification together into one workspace, enabling agents to build software, conduct deep research, automate browser tasks, and collaborate in parallel.
Key Capabilities
- Modes: Provides Work mode for general tasks and Coding mode for repository-aware development.
- Team Structure: Supports a "lead + specialists" agent team, where each agent can be individually configured with models, skills, tools, and permissions.
- Model Access: Bring your own models from roughly 200 hosted, cloud, routed, or local providers, and select them per agent.
- Repository-Aware Coding: Includes code search, symbol relationships, LSP feedback, Git workflows, and reviewable diffs.
- Computer App Control: Agents can operate a desktop application window in the background on Windows or macOS via the accessibility tree, while users can watch a live preview and approve operations item by item.
- WebBridge: Agents can work in the user's real Chrome or Edge through a paired and policy-checked relay, reusing existing login sessions.
- Local Control: Provides inspectable history, permissions, sandboxing, and outbound secret and PII redaction.
- Integration Capabilities: Memory, plugins, MCP, scheduling, and a built-in browser are provided as workspace capabilities.
How to Use You can download the latest desktop version from GitHub Releases, with packages for macOS, Windows, and Linux; the desktop packages include a native Python sidecar. Linux packages should be installed through the system package manager. WebBridge is an optional browser companion distributed separately in the evo-webbridge repository. You can also run from source, which requires Rust and the Tauri CLI, as well as the Tauri prerequisites for the corresponding operating system; use make dev to start the development app (make dev is an alias for make dev-web). EvoFlux is developed and released as a desktop product, and the Web development server is an internal development interface. On first launch, connect a model provider, then start a Work session or open a repository in Coding mode.
Use Cases Suitable for AI agent collaboration scenarios that require local-first operation and controllable data, including software development, deep research, browser task automation, and teams that need multi-agent parallel processing of complex work.
Limitations and Notes The README remains high-level, and detailed contracts and guides are located under the documents/ directory, including the documentation index, the Agent Plugins guide, the Coding architecture, and the development guide. The application keeps the main runtime and project state on the user's machine, model providers are replaceable, and the harness handles context, tools, permissions, memory, and verification. Issues and pull requests are welcome, and it is recommended to keep changes focused and include a minimal relevant test run; vulnerabilities should be reported privately through GitHub Security Advisories. EvoFlux is released under the Apache License 2.0.
