Emdash AI Coding Agent: The Open-Source Multi-Agent IDE (YC W26)

What Is Emdash?

Most coding agent tools assume you run one agent at a time. You open Claude Code, give it a task, wait, review, repeat. Emdash rejects that assumption.

Emdash is an open-source desktop app that runs multiple coding agents in parallel, each isolated in its own git worktree. It supports 22 CLI agents (Claude Code, Codex, Amp, Gemini, Cursor, GitHub Copilot, and 16 others), pulls tickets directly from Linear, Jira, or GitHub Issues, and ships with a Kanban board for tracking agent progress across tasks.

Emdash does not run models. It does not host inference. It orchestrates the CLI agents that do. Think of it as a project manager for your coding agents: it creates isolated workspaces, dispatches tasks, tracks progress, and lets you review diffs in one place.

The YC W26 company was founded by Arne Strickmann and Raban von Spiegel in San Francisco. The GitHub repo (generalaction/emdash) has 2.4K stars, 193 forks, 3,108 commits on main, and 81 releases. Licensed MIT. Available on macOS, Windows, and Linux.

Key Specs

How Git Worktree Isolation Works

The core technical insight behind Emdash is git worktrees. When you assign a task to an agent, Emdash does not just spin up a new terminal. It creates a separate git worktree, a full copy of your working tree that shares the same .git directory but lives in its own directory on disk.

This means two agents can modify the same file in the same repo at the same time without conflicts. Agent A rewrites auth.ts in worktree A. Agent B rewrites auth.ts in worktree B. Neither touches your main working directory. When both finish, you review each diff independently and decide which to merge.

The workflow: click "Add Task" to create a worktree, assign an agent, let it run. When it finishes, review the diff in Emdash's built-in diff viewer, iterate if needed, and open a PR directly from the app.

All 22 Supported Agents

Emdash auto-detects installed CLI agents on your machine. No manual configuration needed. As of March 2026, these are the supported agents:

The agent list is growing. Emdash's CONTRIBUTING.md documents how to add new providers. Since Emdash invokes CLI agents natively, you always get the latest model versions without waiting for Emdash updates. When Anthropic shipped Opus 4.6, Emdash users could use it in Claude Code immediately.

Core Features

Best-of-N: Run Agents Against Each Other

Best-of-N is Emdash's most distinctive workflow. Give the same task to multiple agents (or the same agent with different models) and pick the best result.

Example: assign "refactor the auth module to use OAuth2" to Claude Code (Opus 4.6), Codex (GPT-5.3), and Gemini (2.5 Pro). Each runs in its own worktree. When all three finish, compare diffs side-by-side. One might have cleaner abstractions. Another might handle edge cases better. A third might be fastest but miss error handling.

The cost multiplier is real. Running 3 agents on the same task costs 3x the tokens. But for critical code paths where a mistake costs more than the tokens, Best-of-N is a strict improvement over single-agent execution.

Remote Development via SSH

Emdash connects to remote machines over SSH/SFTP. You can run agents against codebases that live on cloud VMs, dev servers, or GPU nodes without cloning locally.

Requirements: SSH access to the remote machine, git installed on the remote, and an SSH key loaded into your agent. Emdash stores credentials securely in your OS keychain (macOS Keychain, Windows Credential Manager, or Linux Secret Service).

This matters for teams with large monorepos or GPU-heavy workflows. Instead of cloning a 50GB repo locally, point Emdash at the remote and run agents there. The worktree isolation works the same way, each agent gets its own remote worktree.

Emdash vs Superset vs Conductor

Three tools occupy the "parallel coding agent orchestration" space in 2026. They solve the same problem differently.

Superset is the right choice if you want a minimal terminal wrapper around your existing agents. Emdash is the right choice if you want a full project management layer with issue tracking, diff review, and PR creation. Both are open source.

For a deeper comparison, see our Superset vs Conductor vs Emdash breakdown.

Where Emdash Fits With Morph

Emdash orchestrates agents. Those agents need fast, accurate code modifications. That is where Morph fits.

Morph is a fast-apply model that takes a code file and an edit instruction and returns the modified file. It runs at 10,500 tok/s, 25x faster than frontier models on the same task. Coding agents like Claude Code, Codex, and Cursor use fast-apply models under the hood to turn LLM suggestions into actual file edits.

When you run 5 agents in parallel in Emdash, each agent is making dozens of file edits per task. The speed of each edit compounds. A fast-apply layer that is 25x faster means each agent finishes sooner, which means your parallel pipeline completes sooner. Emdash handles the orchestration. Morph handles the execution speed underneath.

Installation

Emdash is available on all three desktop platforms.

brew install --cask emdash

For Windows and Linux, download the installer from the GitHub releases page. Windows ships as MSI and portable EXE. Linux ships as AppImage and .deb.

After installing, Emdash auto-detects which coding agent CLIs you have installed. No configuration file needed. Open a repo, click "Add Task," and select your agent.

Frequently Asked Questions

What is Emdash?

Emdash is an open-source desktop application (YC W26) for running multiple coding agents in parallel. Each agent works in an isolated git worktree. It supports 22 CLI agents including Claude Code, Codex, Amp, Gemini, and Cursor. Available on macOS, Windows, and Linux under the MIT license.

Is Emdash free?

Yes. Emdash itself is free and open-source. You pay for the underlying coding agents and their API keys. The founders have committed to keeping the UI layer permanently open-source.

How does Emdash isolate agents?

Git worktrees. Each agent task gets its own worktree, a full copy of the working tree that shares the same .git directory. Two agents can edit the same file simultaneously without conflicts. When done, you review each diff independently.

Does Emdash send my code to its servers?

No. Emdash stores all app state in a local SQLite database. It does not send code or chat transcripts to any Emdash servers. When you use a coding agent through Emdash, your code goes to that agent's provider (Anthropic, OpenAI, etc.), as it would if you ran the agent directly. Anonymous telemetry can be disabled via settings or the TELEMETRY_ENABLED=false environment variable.

What is Best-of-N?

A workflow where you assign the same task to multiple agents (or models) and compare results side-by-side. Each runs in its own worktree. You pick the best output. Useful for high-stakes changes, model evaluation, and ambiguous tasks with multiple valid solutions.

How does Emdash compare to Cursor?

Cursor is an AI-native IDE. Emdash is an agent orchestration layer. Cursor is one of the 22 agents Emdash supports. You would use Emdash to run Cursor alongside Claude Code and Codex simultaneously on different tasks, or on the same task to compare results.