Open Forge
A practical look at the desktop tool I am building for AI-assisted development.
Open Forge exists because AI-assisted development is often a fragmented experience. We have powerful models that can write code, but we still spend too much time jumping between terminal windows, PR pages, task trackers, and scratchpads. I want one place where that work can stay together.
The problem with fragmented context
When working with AI agents, context is everything. In a standard setup, that context is scattered. You find a bug in Jira, create a branch in your terminal, prompt an AI in a browser or IDE extension, and then manually track what it changed. If you work on multiple tasks, your local environment quickly becomes a mess of half-finished changes and forgotten branches.
This fragmentation doesn’t just slow you down. It makes it harder to review what the AI actually did, leading to a loss of repo hygiene and confidence in the output.
How Open Forge solves it
I wanted a tool that treats AI-generated code like any other engineering work: with clear boundaries and visibility. Open Forge is a desktop application that manages the entire lifecycle of a task.
The product is built around a simple loop:
- Kanban Task Management: Work starts as a task on a board, with GitHub and Jira kept in the same loop.
- Isolated Git Worktrees: When you start a task, Open Forge creates an isolated git worktree. This keeps each task in its own directory with its own branch, so your main environment stays clean.
- Embedded Agent Sessions: AI agents run directly within the app through embedded terminals, keeping the execution context next to the task description.
- Review and Feedback Flow: Before any code is merged, there’s a review step. You can look at diffs, leave inline comments, and send feedback back to the agent in a tight loop.
- Local Voice Input: For a more natural way to provide context or instructions, I integrated local Whisper-based speech recognition for voice-to-text input.
By mapping these features to a single interface, the workflow becomes much calmer. You start a task, watch the progress, review the diff, and move on.
Building the forge
Building Open Forge has meant bridging high-level task management with lower-level system work. I chose a stack that balances a modern UI with the performance needed for file system and process management.
The frontend uses Svelte 5 and TypeScript, styled with Tailwind.
The heavy lifting happens in the desktop layer using Tauri and Rust. That is where the git worktree operations, agent process management, and local SQLite state fit naturally. It is also where the local Whisper-based voice input belongs.
Current state
It’s early, but it’s real. I’m shipping pre-1.0 releases while tightening the core flows.
It’s the most interesting thing I’m building right now. I’m trying to make AI-assisted development feel less like a series of disconnected hacks and more like normal engineering work.