Run Parallel AI Agent Tasks, Then Review Every Change in One Workspace
You've probably tried running an AI coding agent in a terminal by now. It works fine for one task. But what happens when you've got three things you want to try at once, each on a different branch, and you need to actually read through what the agents produced before anything gets merged? Terminal TUIs start to feel cramped fast. Kandev is an open-source tool built around that exact problem: run many agent tasks in parallel, then review all the output in a single integrated workspace.
What It Does
Kandev is a task orchestration and review environment for AI coding agents. You organize work across kanban and pipeline views, assign agents from any supported provider, and execute multiple tasks at the same time. When the agents finish (or while they're still running), you review their output in an IDE-like workspace that bundles a file editor with LSP support, a file tree, a terminal, a browser preview, and a git changes panel into one place.
It's distributed as a native Go binary per platform, with the compiled web frontend embedded directly inside that binary. That's a deliberate architectural choice: the application server doesn't need Node.js, a separate web server, or a frontend build at runtime. Homebrew, Scoop, release archives, and the desktop app all run this native binary. The npm/npx package wraps it with a small Node.js platform selector, so you'll need Node.js if you launch through npm—but not for the server itself.
The multi-provider support is broad: Claude Code, Codex, GitHub Copilot, Gemini CLI, Amp, Auggie, OpenCode, Cursor, Devin, Qwen, Factory Droid, iFlow, Kilocode, Pi, Kimi, AWS Kiro, Qoder, Trae, Oh My Pi, Grok, Hermes, and Antigravity.
Why It's Cool
-
The review problem is the real problem. The README makes a point that lands: "Terminal agent TUIs are great for running agents, but reviewing and iterating on changes there doesn't scale." Anyone who's piped a large diff through a terminal knows this. Kandev treats review as a first-class part of the workflow, not an afterthought.
-
Agentic workflows mix providers per step. You can build multi-step pipelines that use different agents at different stages—Claude Code Opus to design a plan, GitHub Copilot Sonnet to implement it, Codex GPT 5.4 to review the changes. That's a genuinely useful pattern, since different models have different strengths, and it means you're not locked into one vendor's opinion about how a task should be done.
-
Sub-tasks that resume from the parent session. When a task gets too big, agents can spawn sub-tasks that pick up from the parent task's session. The README calls out two concrete uses: splitting an oversized task, or producing several PRs from the same starting point. That's a thoughtful design detail.
-
Automations with real configuration. You can trigger agent tasks on schedules or webhooks, with configurable run destinations, context, and concurrency. This is the difference between a toy and something you'd actually wire into a team's process.
-
Self-hosting and remote access are taken seriously. You can run it locally or on your own infrastructure, and there's a documented path to connect through Tailscale, Cloudflare Tunnel, or another private VPN. The stated goal is a single control plane—offload heavy execution to servers and orchestrate from anywhere, including your phone.
-
No telemetry, not tied to any cloud. Worth noting plainly, since it's increasingly rare.
How to Try It
The README points to Homebrew, Scoop, release archives, the desktop app, and npm/npx as distribution paths. Since the exact install commands aren't spelled out in the README excerpt, start with the repository's release page for your platform.
- Head to the repo: https://github.com/kdlbs/kandev
- Pick your install path. On macOS, Homebrew is the natural fit. On Windows, Scoop. Otherwise grab a release archive or the desktop app. If you prefer npm, the npx package works but requires Node.js.
- Once it's running, the binary serves both the web UI and the API—no separate frontend build needed.
- If you want to connect from your phone or another machine, follow the mobile remote-access guide for Tailscale, Cloudflare Tunnel, or a private VPN.
From there, the docs cover the parts you'll want next: features, workflow tips, running as a service, and architecture. There's also a Discord if you hit a wall.
Final Thoughts
Kandev is clearly aimed at power users—the README says so directly, and the feature set backs it up. If you're running one agent in one terminal and that's working for you, you probably don't need this yet. But if you've started juggling multiple agents, or you're the person on your team who ends up reviewing everyone's AI-generated diffs, the review-first design and the single-workspace approach are worth a look. The multi-provider workflow pipelines are the most interesting part to me, and they're the kind of thing you only appreciate once you've tried to coordinate three different agents by hand.
Follow @githubprojects for more developer tools and open source projects.