Alternatives · Terragon
Migrating from Terragon
A cloud-VM background-agent orchestrator for Claude Code and Codex, wound down within a year of most people hearing about it.
What happened
January 11, 2026
Terragon announced its wind-down on January 11, 2026. The service itself stopped running on February 9, 2026, and the team open-sourced the codebase as a final snapshot rather than leaving it to go quiet.
What Terragon got right
- No local hardware requirement: every task ran on a VM Terragon provisioned, not your laptop.
- Cloud execution per task, so a long-running agent didn't tie up a machine you needed for something else.
What carries over
The same ideas, different names
Nothing here is a new concept. It's the thing you already understood, running under a name that makes sense once more than one person and more than one machine are involved.
- A cloud VM Terragon spun up per task
- A per-task sandbox on a device you register, your own machine or a cloud instance you control
- Terragon's task queue
- A workflow stage with a server-enforced gate
- The agent's cloud runtime, billed by Terragon
- A session on a device you own, with its own per-task and per-model cost line
- Terragon covering the compute bill itself
- Your own devices doing the work, so the economics are yours to size rather than a vendor's to absorb and eventually stop absorbing
What changes once the board is server-side
- You bring the device, your own machine or a cloud instance you provision yourself, so the compute cost is yours to size rather than a vendor's line item that can vanish with the vendor.
- Staffing happens server-side across every device you register, not a single queue against one vendor's own VMs.
- Every task carries its own cost, broken out by stage and by model, with a cap you can set before a task runs rather than a bill you find out about after.
- A task can require a second person's approval before it merges, the separation of duties a single always-on cloud queue doesn't need until more than one person is using it.
Your first week
From Terragon to a merged pull request
- 1
Get started
Sign in and create a workspace. There's no separate account to set up later for the CLI.
- 2
Follow the quickstart
Install the ch CLI, connect the repo you were already working in, and file one task end to end.
- 3
Add the machine you already have
Register it as a device, then check what an agent running on it can reach before handing it real work.
- 4
Learn the workflow stages
See how a task moves from plan to code to review to merge, and where each gate actually sits.
- 5
Check the numbers, then point it at your agent
See what a task costs by stage and model, then run the same agent CLI you were already using.
Questions
What people migrating from Terragon ask
Do I have to give up my own machine?
No, and you don't have to replace Terragon's cloud VMs with your own hardware either. Register whatever you want tasks to run on, a laptop, a spare workstation, or a cloud instance you provision yourself, and CodeHerder treats it the same way.
Does it run the same agent CLI?
Yes. Claude Code and Codex both work as CodeHerder harnesses, alongside Cursor, OpenCode, and Pi, so the agent Terragon was running for you keeps running, just on infrastructure you control.
What does it cost per task?
You bring your own model provider key, so token spend goes straight to your provider rather than through a CodeHerder markup. The platform itself is metered on concurrent agents; see the pricing and calculator pages for real numbers.
Do I need to keep paying for cloud VMs?
Only if you want cloud execution. A registered device can be a machine you already own just as easily as a cloud instance, so the always-on VM bill Terragon used to carry becomes a choice instead of the only option.
Sources
Verified against Terragon's own published notice as of 2026-08-08.
- Terragon shutdown notice
See every tool that's still actively maintained on /compare, or estimate your own cost on the calculator and pricing pages.
Round up your herd.
Bring every human and every agent onto one table. Watch the work move. Costs update as it happens.