Tools like Claude Code, Codex, or OpenCode are great for individual tasks. However, when you need to run multiple AI agents in parallel on real projects, a new barrier arises: coordination and process management issues.
Tools like Claude Code, Codex, or OpenCode are great for individual tasks. However, when you need to run multiple AI agents in parallel on real projects, a new barrier arises: coordination and process management issues.
The main complexity of modern development has shifted from code generation to process management: coordinating multiple AI agents, isolating parallel changes in Git, transferring runtime context, seamlessly processing PR, and combining models from different vendors.
To solve these problems, an experienced Ukrainian developer created Warpforge — an open-source and local-first Agentic Development Environment (ADE).
Warpforge is not another "wrapper" over the model or a new coding agent. It is an operational layer for managing tools already installed in the system (Claude Code, Codex, OpenCode, Qwen Code, Goose and other ACP-compatible agents).
The project combines agent sessions with what is usually "scattered" between IDEs, terminals, and other services: git worktrees, file and diff work, permission control, local services, Kubernetes port-forwards, logs, and final commit and push.
Instead of an endless list of chats, Warpforge focuses on the developer's attention. It is clearly visible what tasks are being performed, what is blocked, where a user response is needed, and what is already waiting for a human review.
Free delegation via Lead Agent: The lead agent can delegate limited tasks to subordinate sub-agents directly during dialogue (for example, when Claude is running Codex or vice versa).
Configurable Workflows: Automated pipeline according to the scheme plan → implement → review ⇄ fix. Each stage can be assigned a separate agent and model (including multiple reviewers).
In this case, the transitions between stages are managed by the local daemon, not the manager-LLM. The pipeline does not commit anything automatically — any result first goes to the Needs review section.
Local-first: All data and processes remain on the user's machine.
Native CLI subscriptions: Warpforge runs already authorized CLI tools, so the user does not need to purchase additional API keys or subscriptions (even a convenient on-the-fly account switcher has been added).
Independent daemon: Even if you close the user interface, the background daemon will continue to execute tasks.
The project is built as a Tauri/React client in conjunction with a Rust/Tokio daemon using SQLite and a local WebSocket API.
Interestingly, Warpforge was originally developed as a tool for managing complex local environments: dependent services, environment configuration, port forwarding and logs for running multiple backends and frontends simultaneously. Working with AI agents and git worktrees grew organically on top of this powerful runtime layer.
The project is completely free and distributed under the free MIT license.
There is currently a stable version for macOS (Apple Silicon). Builds for Windows and Linux are available as untested previews.
The project is open to the community on GitHub .
