Per-project workspaces
Keep a board, terminal, panes and cost records for each repository, with the workspace stored in `~/.devpit` rather than in the project tree.
devpit
devpit is a native desktop workspace for managing coding agents across software projects, combining project terminals, boards, chat, diffs and execution costs. It currently drives Claude Code directly and runs on Linux, macOS and Windows as a free, open-source application.
A desktop control surface for coding agents
devpit is a native desktop workspace for controlling coding agents across one or more software projects. It combines a project terminal, board, chat, file and diff views, cost information and an optional orchestrator in one application, so agent-driven work can be started and reviewed without switching among several tools.
Each project has its own board, panes, terminal target and workspace state. Board steps can run an agent, session or command and then stop with their output, exit code and cost recorded; moving the card forward remains manual unless a lane is configured to move it. devpit keeps this state in `~/.devpit`, separate from the repository, and uses tmux or psmux to keep terminal sessions alive. It currently drives Claude Code through the installed CLI, with additional agent integrations and status hooks documented for other tools.
Core features
Board-driven workflows
Use columns and cards to organize work. A card can run an agent, session or command when it reaches a step, and each execution records output, exit code and actual cost.
Persistent project terminals
Run a real pty behind tmux, or psmux on Windows. The target terminal follows the active project, while sessions continue running after the devpit window closes.
Call and run cost visibility
See the cost of individual chat turns and step executions, including tokens and model details where available. A column can define a spending ceiling for calls.
Cross-project orchestration
Link projects to an orchestrator chat that can inspect their boards, start a session from a card and report back. The orchestrator drafts replies but does not send them without your action.
Integrated review panes
Review files, diffs and a project's served web page beside the terminal. Built-in project capabilities include Notes and Excalidraw, opened as panes.
Practical use cases
Manage agent-assisted feature work
Create a project board with columns and cards, attach agent or command steps, and review each run's output, exit status and cost before deciding whether to move the work forward.
Coordinate parallel projects
Keep several repositories active at once while preserving a separate terminal, board and workspace for each. Switching projects changes the attached terminal without ending the previous session.
Review implementation in context
Inspect an agent's changed files and diffs, open the page served by the project, and use Notes or Excalidraw in adjacent panes during implementation or review.
Triage work across projects
Link multiple project boards to an orchestrator, give it a card through chat and receive a report when a session starts or needs attention, while retaining control over replies and approvals.
Resume interrupted desktop sessions
Leave long-running agent sessions in tmux, close the desktop window and return later to the active process and its project workspace instead of restarting the work.
Pros and Cons
Pros
- Combines project boards, persistent terminals, agent chat, diffs and cost tracking in one native desktop application.
- Keeps workspace metadata outside the repository, avoiding extra files in the project or changes that need to be ignored.
- Makes agent execution more inspectable by recording step output, exit codes and real costs, with configurable spending ceilings.
- Supports manual control and explicit approval points rather than advancing every board step automatically.
- Runs on Linux, macOS and Windows and is free and open source under Apache 2.0.
Cons
- The home page identifies Claude Code as the main agent driven directly today; support for other agents varies between direct control, status reporting and terminal use.
- The workflow is organized around local project workspaces and terminal processes, so teams looking for a hosted, shared workspace may need to evaluate the documented remote and account behavior separately.
- The product is presented as an early release, and some capabilities or integrations are still described as developing.
FAQ
Which coding agents does devpit support?
devpit drives Claude Code through the Claude CLI already installed on your machine. Other agents can report status through the devpit-agent hook, and Gemini CLI works with no additional setup. The documentation also lists Codex, Cursor and other CLIs as supported or terminal-based providers, but the home page identifies Claude Code as the primary supported agent today.
How is devpit installed, and does it run on Windows?
On Linux and macOS, installation is provided through the command `curl -fsSL https://devpit.app/install.sh | sh`. Windows has a per-user installer that includes psmux, the Windows equivalent used for terminal persistence. The Linux build also runs in WSL 2.
Is devpit free, and does it require a separate API key?
Yes. devpit is free and open source under the Apache 2.0 license. It uses an existing Claude Code plan and does not require new API keys for that workflow.
Does devpit modify or add files to my repository?
No. Project boards, panes, terminal state and cost information are stored in `~/.devpit`, while the repository remains separate. Worktrees may be created for lines of work when a checkout is needed, and they are not removed while they contain uncommitted changes.
What control does the orchestrator have?
The orchestrator can read the boards you link, start a session from a card and report back. It does not move cards into lanes that run steps, act on a timer or approve actions in another session. Drafted replies are sent only when you explicitly send them.
Quick Facts
- Category
- Developer Tool; AI Coding Assistant
- Platforms
- Linux, macOS and Windows; Linux also runs in WSL 2
- Primary agent
- Claude Code through the installed claude CLI
- License
- Free and open source under Apache 2.0
- Workspace storage
- Local project state is stored in ~/.devpit, outside the repository
- Architecture
- Rust and Tauri desktop app with tmux-backed terminals; psmux on Windows
Alternatives à devpit
SonOf
SonOf connects to your repo and PM tool, audits the codebase and surrounding product context, and turns approved work into shipped tickets with senior engineering review. It is aimed at founders and engineering leaders who need backlog help without hiring a full team immediately.
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
hob
hob is an independent workspace for coding agents that keeps agent sessions, terminals, history, and follow-up work organized around the tools and providers you already use. It is aimed at developers who want local control over routing, history, and workspace structure rather than a bundled model stack.
Manta AI
Manta AI is an autonomous web app testing tool for teams that want to map application behavior, catch regressions, and generate tests without writing scripts or maintaining selectors. It works from a URL and supports plain-English test flows, run results with screenshots, and scheduled or deployment-triggered checks.