ADE
ADE is a local-first, free and open-source agent development environment that lets developers run multiple AI coding agents in a single workspace
In-depth Report
-
ADE is a native-first, open source and free agent development environment launched in February 2026 by independent developer Arul Sharma. Different from the traditional IDE's idea of "stuffing the agent into the sidebar of the editor", ADE does the opposite - taking the agent as the core and building the entire development workspace around it. It supports mainstream AI coding agents such as Claude Code, Codex, Cursor, Factory Droid, and OpenCode to run in parallel in the same workspace. Each task is independently assigned a Git worktree, and the agents do not interfere with each other. It attracted attention in the developer community as soon as it was released, and is regarded as a very representative product in the emerging category of "Agentic IDE".
-
ADE is developed and maintained by Arul Sharma personally and adopts the AGPL-3.0 open source license. The project was created on GitHub in February 2026 and is still under active development. The author clearly stated in the project introduction that it is "Free forever" and there is no plan for commercialization - at least at this stage, this is an open source project driven purely by developers. ADE's product positioning is very clear: it is not another AI coding assistant, but an "intelligent agent operating platform." You don't need to replace your accustomed agents - Claude Code continues to write the backend, Codex handles the frontend, and Cursor does refactoring - they all run in ADE, sharing isolated instances of the worktree in the same code base. This "agent-independent" neutral attitude is the most essential difference between it and competing products such as Cursor and Warp.
-
Parallel Worktree The core capability of ADE is Git worktree. Each time a task is started, ADE automatically creates an independent worktree - containing its own branch, working directory, terminal, port and history. Multiple agents can work on different branches of the same warehouse at the same time without code conflicts. Worktrees support stacking, and upstream branches can be automatically rebased after merging. The actual meaning of this design is: you no longer need to "wait a minute, I will stash the current branch first and then let the AI change other things." In ADE, developers can simultaneously assign Claude Code to fix a bug, Codex to add a new API, and Cursor to write unit tests. The three agents advance in parallel without knowing each other. Intelligent plug and play ADE currently supports five mainstream AI coding agents: Claude Code (Anthropic), Codex CLI (OpenAI), Cursor, Factory Droid, and OpenCode. Add API Key or subscribe in Settings to access. Every time you initiate a task in Chat, you can choose which agent to use for the current task - all agents share the same worktree context and display diff and approval gates in real time. Built-in PR management The PR submitted by the agent will be directly displayed inside ADE - diff comparison, CI status, comments, and merge buttons are all available without the need to open the browser and switch to GitHub Tab. Supports automatic merge (green merge), and also has specialized management of stacked PR flows. Sync across devices ADE offers a native desktop client for macOS, a CLI (ade terminal command), and an iOS companion app (distributed through TestFlight). For tasks started on Mac, you can use your mobile phone to approve diffs and check the status. All data is synchronized in real time through the self-built synchronization layer, and the operating environment of the same machine runs behind it. CTO, Commander-in-Chief of Intelligence ADE has a built-in resident agent named CTO - it has a global context across all worktrees, can pull tasks from Linear, delegate them to specific execution agents, and report the results after completion. For complex tasks that require multi-step collaboration, the CTO acts as a "project manager."Built-in toolset The editor has 35+ tools built-in such as browser, file management, Git operation, iOS simulator, etc., and the agent can call it directly. In addition, over 50 agent-controllable actions are exposed through the MCP service - including splitting the screen, launching a child agent, starting the development server, opening the browser, navigating files, and switching projects.
-
ADE is completely free and open source (AGPL-3.0). Users only need to bring their own API Key or subscription to the AI agent (Claude Code / Codex / Cursor each have a payment model). ADE itself does not charge any platform fees. There are currently no firm plans for an enterprise version, and there are no paid plans for cloud hosting. This "platform is free and the agent pays for itself" model makes ADE more like an open source project at the infrastructure layer rather than a commercial SaaS product.
-
Among the developer community, word of mouth about ADE is generally positive. Most users appreciate its product philosophy of “taking the agent as the protagonist” and its engineering implementation of worktree isolation. Positive comments focus on: "Finally no need to cut fancy tabs" - parallel worktree is the most mentioned highlight by users; "Agent independence is so comfortable" - not locked into a certain company's agents; "Local priority is reassuring" - the code and running status are all local; "native macOS experience is smooth" - there is a clear perception difference compared to Electron applications. Criticisms and limitations: Currently, there is only a macOS version, and the Windows version is in a suspended state; initial configuration requires certain technical thresholds (Git operations, API keys management); the documentation is still incomplete, and some advanced functions lack written guidance; the iOS side has limited functions, which can only be approved and viewed at present, and new tasks cannot be initiated; as a personal open source project, the sustainability of long-term maintenance and iteration remains to be seen.
-
The timing of ADE's emergence is interesting. In the first half of 2026, competition on the “Agentic IDE” track has become increasingly fierce. JetBrains launched Air in March (free preview, based on the discontinued Fleet code base), Warp launched universal agent support in April (using the terminal as the main interface of ADE), and Cursor continues to delve into agent collaboration within the editor. In such a crowded track, ADE has won its own ecological niche with its two differentiated positionings of "agent neutrality" and "local priority". Some industry media commented that ADE represents a paradigm shift signal "from IDE to ADE": not an enhanced editor like Cursor or Copilot, but a completely new category - a workspace built around agents, not around cursors. Of course there are also cautious voices. Comments pointed out that ADE currently has only one full-time developer (Arul Sharma) and that the AGPL-3.0 protocol is also restrictive in a commercial environment. Comparing JetBrains Air (backed by JetBrains company resources) and Warp (which has completed multiple rounds of financing), there are variables in ADE's long-term competitiveness.
-
There are currently no apparent legal or ethical controversies with ADE. Potential risks worthy of attention include: single point of failure risk - there is only one core maintainer of the project; platform compatibility limitations - macOS-only limits the expansion of the user base; open source agreement restrictions - AGPL-3.0 may be restricted in some enterprise environments.
-
ADE is most suitable for: developers who use multiple AI coding agents in depth on a daily basis; full-stack or back-end developers who need to handle a large number of parallel tasks at the same time; teams that are concerned about code security and privacy and are unwilling to upload code to the cloud; macOS users. Not suitable for: Windows users (no version yet); light AI coding users (the learning curve may not be proportional to the benefits); users who need a complete IDE debugging experience (ADE is positioned at the orchestration layer, not an editor replacement). Alternatives: JetBrains Air (suitable for JetBrains ecosystem users), Warp (suitable for heavy terminal users), Cursor (suitable for in-editor experience), GitHub Copilot (low entry threshold).
-
ADE is a highly recognizable presence in the Agentic IDE wave of 2026. It accurately captures the trend that "agents are becoming the protagonist of the development process" and enters the market with a local-first, agent-neutral attitude. For developers already working with multiple AI-coded agents, ADE provides a unified command center. But there is still a long way to go from personal open source projects to mature products.
User Reviews
-
ZoeyPerry—Tried ADE's parallel worktree approach — it's a blast. Before, fixing a bug in Claude Code meant stashing the current branch first; now I just spin up a new worktree and let it run, zero overhead. -
王瑶悦—Finally a tool that doesn't tie you to a single agent. In ADE you run Claude Code and Codex at the same time, each doing its own thing without getting in each other's way — that's the agent workflow I wanted. -
Hnflo—Open source, free, and local-first — those three alone are worth a star. If you're already using multiple coding agents, ADE saves way more mental effort than you'd expect. -
Judith.CampbellJr10—To be honest, it's still pretty rough right now. The worktree approach itself is fine, but the docs aren't very complete — for some features you have to dig through GitHub issues to figure them out. -
GBell007—As a Windows user, I feel left out. The project page says the Windows version is on hold, and I don't know when I'll get to use it. For now I'm just making do running the macOS version on Parallels. -
郭丹丹—ADE's iOS companion app is really handy — I got a notification while out eating that the Agent had finished, pulled out my phone, glanced at the diff, and merged it right there, sparing me from firing up the computer. -
MestanSözeri—Only three days in, and I threw the refactor of three microservices in the project at three different Agents, each with its own worktree, and got it all done within a day — versus at least a week of fiddling before. -
Bryan30—The only thing I worry about is that it's just Arul working on it full-time; if he ever stops maintaining it, that'd be awkward. I hope the community grows to share the load. -
NicoleVasquez369—Switched over from Cursor. ADE's core thinking really is different: Cursor stuffs the Agent into the editor, while ADE rebuilt the editor around the Agent. After using it, the latter feels like the right call. -
Billy.Jones0077—A native macOS app is just smooth — a totally different feel from those Electron wrappers. The SwiftUI interface is very responsive and snappy. -
MindyReyes—The CTO feature is still a bit weak right now, but the direction is right. Once it matures I won't have to act as project manager and hand out tasks myself. -
Joyce.Young591—The first-time setup took a bit of effort, mainly because you have to enter the API keys one by one. But once configured it's smooth sailing, and the Guest Mode design is thoughtful too. -
TReyes_2021—ADE's PR management is well done — PRs submitted by the Agent can be viewed right inside it with diffs and CI status, no need to switch to the GitHub web page. It also handles stacked PR flows really well. -
greenbird550—The only thing I'm not quite happy with now is that the CLI tool `ade` isn't feature-complete yet. Some operations available on the desktop side can't be found in the CLI. -
Donald877—After I demoed ADE at the office, the Windows crowd all started asking when the Windows version is coming. If it ships, I think team collaboration is going to be really sweet. -
crazyfish235—The local-first positioning is just great — all the code and running state stays on your own machine, so you don't have to worry about the Agent shipping sensitive code off to some weird server. -
yu3xanqqbs—Compared it against JetBrains Air for a round — ADE wins on being lightweight and clean. Air has more features but is way too heavy; ADE hits exactly the balance I want. -
Scott.Stephens_2023—After a week on ADE, my biggest takeaway is 'there's no going back'. The old routine of manually switching branches, opening terminals, and eyeballing diffs now all happens in a single window. -
SJones_66747—The open-source community vibe is great. I filed a feature request on GitHub and the author replied the very next day. Products from indie developers often have this kind of warmth. -
流年_19—IDEs are designed for a single person; ADE is designed for 'one person with a team of agents'. So well put — and that's exactly how it feels in actual use. -
smallswan807—The overall experience is good, but I've never touched half of the 35+ built-in tools. Lots of features, but there's currently no systematic onboarding that tells you what each of them actually does. -
AnthonyJones_2021—My favorite scenario: have Claude Code fix backend bugs, Codex write frontend components, and Cursor do performance optimization — three agents running simultaneously in their own worktrees. This was simply impossible before ADE. -
Cheryl_GomezZ—The MCP service exposes 50+ actions — agents can split panes, start servers, and open browsers on their own. That level of programmability makes me feel it's not just a tool, but more of a platform. -
Grace502—Honestly I'd love a simple web or cloud version — sometimes I'm out and about and want to know if the Agent has finished. Right now the iOS app is the only way to check, and even that depends on the desktop being online. -
SHtay—One afternoon of trying it and I was won over by the worktree isolation design. No more fear of AI-written code messing up my main branch. -
Grace_Scott—Free and open source. AGPL-3.0 does come with more restrictions, but for individual developers it's totally sufficient. Hope Arul can find a sustainable business model. -
掠影_16—Not many people in the Chinese-speaking community seem to know about ADE yet — a tool this good deserves a bigger audience. I wrote a hands-on writeup and shared it with colleagues, and they were all quite interested.