Open source alternative to Conductor
Last verified
Orchestrate multiple coding agents in parallel, each running in an isolated workspace on macOS, Linux, or a remote SSH server.
Run one agent per branch simultaneously, then review each agent's changes in Xum's central git status view before deciding which approach to merge. The worktree runtime keeps branches independent so agents cannot overwrite each other.
You only run one agent at a time. Standard Claude Code or a single terminal session is simpler when there is no parallelism to manage.
Point two or three workspaces at the same codebase with different models, run the same prompt, and compare their outputs using Xum's git divergence view. Useful for benchmarking model quality on your actual codebase before committing to a provider.
You have a fixed model contract and no interest in multi-model comparison. The overhead of managing multiple workspaces does not pay off for a single-model workflow.
Xum's server mode exposes a responsive web UI accessible from a phone or tablet. Start a fleet of agents on your workstation, then check their status and review output from a mobile browser without installing anything on the mobile device.
You only use Xum at your desktop. Server mode adds a hosting step that is not worth the setup for local-only workflows.
The SSH runtime runs each workspace on a remote server you control, keeping source code off local machines. This suits teams with centralized infrastructure policies who want agents executing on the server rather than individual laptops.
Your team already uses Coder's remote development tool for workspace management. The overlap may make native Coder integration a better fit than a separate Xum installation.
Running multiple coding agents on the same codebase exposes a gap in existing tools: agents share the filesystem, overwrite each other's edits, and their outputs are scattered across separate terminal windows with no unified way to track what each one actually changed. Reviewing three agents' work means switching between sessions, manually comparing branches, and discovering conflicts only after the fact.
The core problem is coordination overhead. Each agent may produce good code independently, but without isolated workspaces and a shared status view, the developer spends as much time managing the agents as evaluating their output. Scaling to five or more concurrent agents with standard CLI tools quickly reaches a point where coordination cost erases the productivity gain from parallelism.
Local, Worktree, and SSH runtimes give three levels of workspace isolation. Local runs the agent in your project directory. Worktree creates a git worktree per agent so each one works on a separate branch copy. SSH runs agents on a remote server you control, keeping source code off local machines.
Connect Claude (sonnet-4-*, opus-4-*), GPT (gpt-5-*, gpt-6-*), Grok, any OpenRouter model, or a local Ollama instance. Model selection is per-workspace, so you can run a faster model on one branch and a more capable one on another at the same time.
The main dashboard shows every agent's git status and branch divergence at a glance. As agents commit changes, you see what each modified without switching windows or running manual git commands, which speeds up code review and merge planning.
Xum's agent loop includes Plan/Exec mode, vim-style inputs, and a full keyboard shortcut reference. The interface is designed for keyboard-first workflows to manage a large fleet of agents without reaching for the mouse.
Agents' proposals and plans render as formatted markdown with Mermaid diagrams and LaTeX math. This makes it practical to review a complex refactoring plan or algorithm description directly in Xum's output panel without exporting to another tool.
A costs tab shows each workspace's token consumption and estimated spend in real time. When running multiple models simultaneously, you see which workspaces are consuming the most context before budgets spike unexpectedly.
Xum and Conductor both offer parallel coding agent management with isolated workspaces, but they represent different deployment models. Conductor was a managed commercial service; Xum is self-hosted and open source.
Conductor shuts down on 2026-11-20. Teams that built workflows on Conductor should plan their migration before that date.
| Feature | Xum | Conductor |
|---|---|---|
| License | AGPL-3.0 | Proprietary |
| Self-hosting | Yes | No |
| LLM providers | Claude, GPT, Grok, Ollama, OpenRouter | Proprietary selection |
| Pricing | Free (self-hosted) | Subscription |
| Platforms | macOS, Linux, SSH server | Managed service |
Xum's parallel-agent model mirrors Conductor's core workflow: each agent runs in an isolated workspace, and you review changes across agents in a central view. The migration path is conceptually straightforward, with Xum adding LLM flexibility that Conductor did not offer. Conductor users who relied on GPT or Claude can keep those models; those who want to test local Ollama models for cost control can do that too.
The practical differences are infrastructure and ownership. With Xum, source code stays on your machine or server. There is no vendor to call when something breaks, which means your team owns both the upside (full configuration control) and the downside (you manage the binary and updates). Conductor's managed experience meant less operational overhead, but that trade-off disappears when the service ends.
For teams evaluating alternatives now, Xum is the strongest direct replacement in the parallel coding agent category: it matches the core workflow, runs where you need it, and carries no ongoing subscription cost.
Download the pre-built binary for macOS or Linux from GitHub Releases: https://github.com/coder/xum/releases/latestYes. Xum is free to download and self-host. The binary is available from GitHub Releases for macOS and Linux with no license key or subscription. The AGPL-3.0 license permits commercial internal use without restriction. The only limitation applies if you offer Xum itself as a hosted service to others, in which case you must distribute your source modifications.
Xum supports Claude (sonnet-4-* and opus-4-), GPT (gpt-5- and gpt-6-*), Grok, and any model via OpenRouter. Ollama is also supported for running LLMs locally without an API key. Model selection is per-workspace, so different agents can use different models in the same session.
The project was originally called Mux, a common technical abbreviation for multiplexer. Mux.com raised a trademark concern, and Coder Technologies chose to rename the project to Xum rather than pursue a dispute. The underlying code and GitHub organization are unchanged; only the name and repository path moved.
Xum ships pre-built binaries for macOS and Linux only. There is no documented Windows installer. Windows users can run Xum via WSL2 or point it at a remote Linux server over SSH.
Local runs the agent directly in your project directory, sharing the filesystem. Worktree creates a git worktree for each agent so each one works on a separate branch copy with no overlap. SSH runs agents on a remote server you connect to, keeping source code off your local machine. Most individual developers start with Worktree; teams with centralized servers use SSH.
Self-hosted AI agent management with org charts and budgets
#1 in the Conductor guide
Drop-in Conductor replacement with 27 or more agents on macOS, Windows, and Linux under MIT
#2 in the Conductor guide
MIT-licensed control plane for 6 or more agents with mobile and web clients and a large community
#3 in the Conductor guide
Daemon-based architecture for agents running on a VM or Docker host with full mobile parity
Sponsors
Alternative to
Superhuman
AI email assistant that drafts replies and clears your inbox
Alternative to
Grok Bot
AI teammates you own: your keys, your model, your machine.
Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.
The projects we add and every new guide. Free, unsubscribe any time.
Claude Code to 
OpenCode
Jira to 
Windshift
Notion to 
AppFlowy
Slack to 
Mattermost
Linear to 
GitLab
Wispr Flow to 
Handy
Arc to 
Zen Browser
CapCut to 
Concat
Grok Bot to 
rakazo
Superhuman to 
Inbox Zero
Airtable to 
AITable
Muse to 
OpenClaw
Railway to 
Coolify
GitHub Copilot to 
Cline
Datadog to 
NetdataConfluence to 
Wiki.js
Microsoft Teams to 
Rocket.Chat
Firebase to 
Supabase
Pinecone to 
Qdrant
Zapier to 
n8n
1Password to 
KeePassXC
Devin AI to 
OpenHands
Asana to 
Leantime
Contentful to 
Strapi
