Proliferate lets you run every AI coding agent in parallel, in one workspace

The AI coding agent space has become a crowded battlefield. Between Claude Code, Codex, OpenCode, Grok, and a dozen others, developers are often forced to pick a single tool — or juggle multiple terminals, each with its own project state, history, and quirks. Proliferate, a new open-source project, aims to end that by running all of them in parallel, in one unified workspace.

The core idea is simple: each task gets its own isolated git worktree, with a dedicated branch, terminal, conversation context, and review state. The Proliferate desktop app then runs each agent through its native harness — meaning Claude Code runs like Claude Code, Codex runs like Codex — without any reconfiguration or loss of features.

For developers, the appeal is obvious. Instead of committing to one agent's strengths and weaknesses, you can let them compete on the same task, review their separate diffs, and merge the best result. It also solves a practical pain point: running multiple agents in the same working directory is a recipe for conflicts. Isolated worktrees make parallel execution safe by construction.

The control plane is self-hostable, with deployment docs covering Docker, AWS, GCP, Azure, Kubernetes, and even air-gapped operation. That's a strong signal for teams with compliance or data-residency requirements — you're not forced to route your code through a third-party cloud. The desktop app can point at a remote control plane, so coordination can live on a shared server while agents run on individual machines.

Local development requires Python 3.12+, uv, and Docker for the database, and it supports named dev profiles to keep multiple worktrees from colliding on ports or generated Tauri config. The project is AGPL-3.0 licensed, which will please open-source purists but might give pause to enterprises that prefer permissive licenses.

Proliferate is an ambitious attempt to tame agent sprawl. It's early days — the project is on GitHub and clearly pre-1.0 — but the architecture is thoughtful, and the self-hostable control plane is a differentiator. If you're tired of choosing between AI coding tools, this is worth a look.

Why it matters

The proliferation of AI coding agents hasn't just created choice — it's created chaos. Each agent has its own context, its own state, and its own way of working, and moving between them is friction. Proliferate's worktree-per-task model is a genuinely useful abstraction: it lets you treat agents as interchangeable workers on a shared project, rather than walled gardens. For teams, the self-hostable control plane also addresses the trust gap that's slowed enterprise adoption of AI coding tools.

It's not a magic bullet — you still need to review the output and reconcile divergent approaches — but it removes the infrastructure excuse for not experimenting with multiple agents. If this matures, it could become the standard way to work with AI coding assistants, the same way CI runners standardized automated testing.