Lich v0.90.0: Multi-Agent IDE with Worktrees and Browser Control

What changed

Lich has released version 0.90.0 of its multi-agent Integrated Development Environment (IDE), designed to facilitate simultaneous work from multiple AI agents. The core functionality allows each agent to operate on its own Git worktree, enabling them to work on separate branches of a project concurrently within a single application window. This approach supports various AI agents, including Claude Code, Codex, and opencode, alongside custom agents.

Key features introduced or enhanced in this version include:

Agent Collaboration and Management

• Agents Side-by-Side: Lich can arrange agent sessions in a visual layout, providing an overview of agent status (busy, waiting), and their current context window usage.
• Worktree per Task: New sessions can be branched off any base, with Lich automatically copying .env files and assigning a dedicated development server port. Uncommitted work can be carried over to these new worktrees.
• Delegation: Agents can delegate tasks to other Lich sessions and receive responses, with similar functionality available via the lich command-line interface.
• Sandboxed Agents: Agents can be run in sandboxed environments that restrict access to sensitive credentials and other repositories, enhancing security.

Interaction and Review

• Review Diff, Comment Lines: Developers can review code changes made by an agent directly alongside its session, add line-specific comments, and submit all feedback as a single prompt.
• Point at the Page, Tell the Agent: Agents can drive a browser instance, allowing developers to interact with the same web page. Users can highlight specific elements, add notes, and send this feedback to the agent as a prompt.
• Pull Requests without Browser: The tool facilitates the creation, inline review, and merging of pull requests directly within the Lich window.

Customization and Usability

• Theming: Lich supports custom themes importable via JSON or Git repositories.
• Dev Servers in Sidebar: Frequently used development servers can be saved and managed from a sidebar for quick start/stop actions.
• Search and Notifications: A Ctrl/Cmd+K shortcut allows quick searching for sessions by name or content. Desktop notifications alert users when an agent is waiting for their input.

Installation is streamlined via a single curl command for Linux, Homebrew for macOS, and installers or portable zip files for Windows. The tool supports a wide array of AI providers, detailing specific feature support for each in a comprehensive table.

Why it matters for builders

Lich v0.90.0 introduces a novel approach to AI-assisted development by enabling multiple AI agents to work in parallel on distinct project branches. This capability directly addresses the complexity of modern software development, allowing builders to leverage different agents for specialized tasks simultaneously. The integration of worktrees, direct browser control for agents, and streamlined code review workflows significantly reduces context switching and manual overhead, making AI agents more integrated and effective collaborators.

Practical impact

Developers can now install Lich and begin experimenting with multi-agent workflows. The immediate impact is the ability to manage several AI coding assistants concurrently, each working on a separate feature branch or bug fix within its own isolated worktree. Builders should explore setting up sessions with their preferred AI providers, such as Claude Code or Codex, and test the inline code review and browser interaction features to see how they can accelerate their debugging and feature development processes. The delegation feature also offers a path to explore more complex agent orchestration scenarios.

Caveats and source limits

The provided source is a GitHub repository description and README. While it details extensive features and installation instructions, it does not include independent benchmark results comparing Lich's performance against other development tools or agent frameworks. Specific details regarding the exact number of concurrent agents supported, performance metrics for complex delegation tasks, or pricing information for any associated AI services are not present. The source also notes that macOS and Windows builds are not yet signed, which may trigger security warnings during installation.