Ensembyte: An Open-Source, Local-First Client for Running AI Coding Agents
Ensembyte is an open-source, local-first desktop client for driving Claude Code, Codex, Cursor and other coding agents. Try it, report issues, or send a pull request. Agar description ka character limit chhota nikle, ye short version use kar: An open-source desktop client for running AI coding agents locally. Feedback, issues and pull requests are welcome.
Ensembyte: An Open-Source, Local-First Client for Running AI Coding Agents
By Soumya Chakraborty
I built Ensembyte, a desktop application for working with AI coding agents from one place. It is open source under the MIT license, and I am looking for people to try it, break it, and tell me what to fix. If you find a bug or have an improvement in mind, please open an issue or send a pull request. If the project is useful to you, a star on GitHub helps other developers find it.
Repository: https://github.com/soumyachk101/Ensembyte
Website: https://ensembyte.vercel.app
What Ensembyte does
Ensembyte does not replace the coding agents you already use. It connects to the command-line agents installed and signed in on your machine, such as Claude Code, Codex, Cursor, OpenCode and Grok, and gives them a shared native interface. A few providers, DeepSeek for example, connect through their native API with a key. Support varies by platform, and the README has the full table.
Your subscriptions stay with you and your code stays on your machine. Ensembyte collects no telemetry and does not need an account.
The interface gives you:
- Project and thread sidebars with search and pinning, and a separate runtime for each thread.
- Live streaming of the agent's reasoning, tool calls, terminal output and file changes in one view.
- Permission and approval prompts that you answer inline, without switching to a terminal.
- A hidden git checkpoint for every model turn, with a diff inspector that lets you accept or reject individual hunks.
- Support for MCP servers over stdio, HTTP and SSE, including an OAuth flow.
- Token spend and rate-limit counters, so provider usage stays visible.
Two native clients
Ensembyte ships as two separate applications, each written in the native toolchain for its platform.
Neither client uses Electron or a webview. The goal was low memory use and responsive rendering on each platform.
Hydra: parallel agents on the macOS client
Hydra is the multi-agent mode in the macOS client. A lead agent plans a task and hands pieces of it to as many as eight parallel heads. Each head works in its own git worktree, so heads cannot overwrite each other, and completed work is merged back with a single review step. While they run, you can watch the status and output of every head.
How the code is organized
Both clients follow the same four-layer structure: Core, Services, UI and App. Dependencies run in one direction only. A file in Core never reaches into a higher layer, and new code goes into the lowest layer that satisfies its dependencies. ARCHITECTURE.md in the repository describes this in detail and is the best place to start before sending a pull request.
Getting started
Prebuilt packages are on the releases page of the repository. macOS gets a signed and notarized DMG, and Windows and Linux get installers or archives. To build from source, follow the README for macOS and Ensembyte-rust/README.md for Windows and Linux.
Install and sign in to at least one supported agent CLI first, since Ensembyte drives the tools you already have.
Current state
Ensembyte is young, and some rough edges are expected. Bugs, glitches and platform-specific problems are the most useful things you can report right now.
How you can help
- Try it with the agent you use every day and report what breaks. A good issue includes your operating system and version, the agent and its CLI version, what you did, and what you expected to happen.
- Send a pull request for fixes and improvements. For larger features, open an issue first so we can agree on the design. Keep each pull request to one concern, and read CONTRIBUTING.md and the code of conduct before you start.
- Tell me which agents or features should come next.
- Star the repository if it is useful to you.
The code is at https://github.com/soumyachk101/Ensembyte. Thank you for taking a look.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article