Terragon alternatives for background coding agents
Published July 10, 2026 · by the AQ team
Terragon shut down in January 2026: terragonlabs.com now shows a shutdown notice, and the team published an open source snapshot of the codebase (github.com/terragon-labs/terragon-oss, January 16, 2026). The closest replacement depends on what you used Terragon for. If your team ran Claude Code and Codex as background agents in the cloud and you want that back on infrastructure you control, AQ is the most direct successor. If you want something free and local and you work on a Mac, Conductor. If open source is the requirement, OpenHands. If you would rather stay inside a vendor's cloud, Cursor cloud agents and GitHub Copilot Agent HQ both run managed background agents. All claims below are as of July 2026.
What Terragon was, and what to look for in a replacement
Terragon was a remote background agent orchestrator: you pointed it at a repository and it ran Claude Code, Codex, and other coding CLIs in the cloud, so tasks kept moving while you did something else. The open source snapshot exists, but it is a snapshot, not a maintained project, so most teams are migrating rather than self-hosting it.
Four questions cover most of the replacement decision:
- Which agents does it run? Some platforms run the CLIs you already subscribe to (Claude Code, Codex, Cursor Agent). Others run their own proprietary agent, which changes your workflows and your billing.
- Where does execution happen? Your own VM, your laptop, or the vendor's cloud. This decides who can reach your code and what happens if the vendor disappears, which is exactly the lesson Terragon taught.
- Can your team see the work? Background agents produce pull requests either way; only some tools let teammates watch and steer the live session before the PR lands.
- How do you pay for models? Some tools proxy inference and mark it up. Others use the per-user subscriptions you already have.
Terragon alternatives at a glance
| Tool | Platform | Execution model | Team features | Price posture |
|---|---|---|---|---|
| Terragon (reference) | Was web-based | Vendor cloud | Shut down January 2026 | Shut down |
| AQ | Browser (macOS, Windows, Linux) | Your own VM or an AQ-managed VM; real CLIs in tmux | Multiplayer workspaces, sharing, guest preview links, Linear intake | $100/user/mo promotional (standard $200), 14-day trial |
| Conductor | Mac desktop app | Local, on your Mac | Single-player | Free |
| OpenHands | Web, or self-hosted | Their cloud or your own infrastructure (enterprise Kubernetes runtime) | Team and enterprise features | Open source; paid cloud |
| Warp Oz | Via Warp | Warp's cloud, fleets of agents | Team-oriented | Commercial (Warp plans) |
| Cursor cloud agents | Inside Cursor | Dedicated VMs in Cursor's cloud | Cursor team plans | Bundled with Cursor plans |
| GitHub Copilot Agent HQ | GitHub and IDEs | GitHub's cloud | Org-wide, mission control view | Included with Copilot plans |
1. AQ: Terragon's model, on your own infrastructure, with your team
AQ is the multiplayer workspace where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud.
Like Terragon, AQ runs the coding CLIs you already use (Claude Code, Codex, Cursor Agent, Kimi, Grok, and plain shells) rather than a proprietary agent. Each workspace is an isolated git worktree, agents run as real CLIs inside tmux, and the session streams live to the browser, so it works the same from macOS, Windows, or Linux. Two things Terragon never had: execution happens on a VM you own (or an AQ-managed VM (included with the paid plan)), and workspaces are multiplayer, so a teammate can open the same workspace and see the same live terminal, editor, and dev-server preview. Comments pinned on a preview can be sent to the agent as prompts. Linear intake (label an issue "ai-task" and a workspace appears in the sidebar, with status syncing back), per-user GitHub auth, and per-workspace PR tracking cover the team workflow. AQ uses per-user CLI logins and does not proxy or mark up your model usage.
Honest fit: the strongest option if Terragon's shutdown convinced you to keep execution on infrastructure you control, and you work with other people. It is a commercial product with a 14-day trial, not a free tool.
2. Conductor: free and local, if you are on a Mac
Conductor (conductor.build) is a free Mac desktop app that runs parallel Claude Code, Codex, and Cursor agents in isolated workspaces on your Mac. It is polished, well funded (YC-backed, with a $22M Series A), and it has a Linear integration. Honest fit: excellent for a solo developer on a Mac. The limits are structural: Mac-only (Windows is a waitlist as of July 2026), single-player, and local execution, so agents only run while your Mac does. That is a different model from Terragon's cloud background agents.
3. OpenHands: open source and self-hostable
OpenHands (openhands.dev) is an open source platform for cloud coding agents: use their hosted cloud, or run it yourself, up to an enterprise Kubernetes runtime. Honest fit: the safest choice if Terragon made "we can run this ourselves forever" a hard requirement. Note that OpenHands runs its own agent against your model keys rather than wrapping the CLIs you already subscribe to, so workflows built around Claude Code or Codex specifics will change.
4. Warp Oz: cloud agent fleets from the Warp team
Warp Oz (warp.dev/oz) is an orchestration platform for fleets of cloud agents, and it is built with teams in mind. Honest fit: worth evaluating if your team already lives in Warp, or wants a managed, fleet-scale take on background agents and is comfortable with execution in Warp's cloud.
5. Cursor cloud agents: if your team already pays for Cursor
Cursor bundles cloud agents that run on dedicated VMs in Cursor's cloud. Honest fit: the lowest-friction option for existing Cursor teams, since you can hand work to a background agent from the editor you already use. You accept vendor-cloud execution and Cursor's agent rather than a choice of CLIs.
6. GitHub Copilot Agent HQ: background agents inside GitHub
Agent HQ brings agents from multiple vendors into Copilot, with a mission control view for tracking them. Honest fit: the closest thing to "assign work to an agent from an issue or a pull request without adopting new tooling." It is a delegation surface rather than a live terminal you watch and steer.
How to choose
Pick AQ if the shutdown burned you and you want Terragon's workflow back on your own infrastructure, with teammates in the same live workspace. Pick Conductor if you are one person on a Mac and free matters more than cloud execution. Pick OpenHands if open source is non-negotiable. Pick Cursor cloud agents or Copilot Agent HQ if you would rather deepen an existing vendor relationship than adopt a new tool. And apply the 2026 lesson generally: check who maintains and funds whatever you pick. Crystal was deprecated in February 2026, and vibe-kanban lost its commercial sponsor when Bloop shut down in April 2026. This category is consolidating, and migration costs are real.
Frequently asked questions
Is Terragon coming back?
There is no indication of that as of July 2026. terragonlabs.com shows a shutdown notice, and the team's last public artifact is the open source snapshot published on January 16, 2026.
Can I keep running Terragon from the open source snapshot?
Technically yes: the code at github.com/terragon-labs/terragon-oss can be self-hosted. But it is a snapshot with no maintenance, security updates, or support, so most teams treat it as a reference and migrate to a maintained platform instead.
What is the closest replacement for Terragon's cloud background agents?
For teams, AQ is the most direct successor: the same idea of running Claude Code, Codex, and other CLIs remotely, but on a VM you control and with multiplayer workspaces. If you prefer a managed vendor cloud, Cursor cloud agents and GitHub Copilot Agent HQ are the simplest options.
Do these alternatives require new model subscriptions?
It varies. AQ uses each user's existing CLI logins (Claude Code, Codex, and others) and does not proxy or mark up model usage. OpenHands needs your model API keys. Cursor and Copilot agents bill through those products' own plans.