Middleman

Middleman is a local-first, open-source platform designed for orchestrating AI agents in coding projects. It addresses the challenge of managing multiple AI agents by introducing a persistent AI manager that handles project management tasks such as dispatching work, tracking progress, and managing merges, allowing users to focus on setting direction and standards rather than micromanaging development workflows.

The platform enables users to create a manager for each coding project by connecting it to a code repository and selecting preferred AI models. Users can onboard the manager with their working preferences, coding standards, and task breakdown strategies, which the manager retains across sessions through persistent memory. Middleman supports parallel execution by spawning multiple worker agents simultaneously, each handling different aspects of a project—such as backend and UI development—in isolated git worktrees. A dedicated merger agent serializes completed work into the main branch, reducing conflicts and the need for manual oversight.

Middleman’s event-driven architecture ensures that the manager is never blocked, enabling it to dispatch tasks, process status updates, and steer agents without waiting on individual workers. The tool supports multi-model teams, allowing the manager to route tasks to the most suitable AI model for each job, such as using Codex for code generation or Opus for UI work. All operations are conducted locally, and the platform is delivered as a self-hosted daemon that runs on the user’s machine, ensuring that code and API keys remain on localhost.

0, and its source code is available for users to review and modify. Middleman is suitable for developers seeking to automate and streamline the management of AI agents in software projects while retaining control over their local environment.