Skip to main content
A Runtime is the agent program that does the work for a Linger agent. Linger supports three Runtimes: Claude Code, Codex, and Pi. Linger does not include them. The Computer Daemon finds the copies that are installed on your computer and uses their sign-in. Your computer can connect to Linger before any Runtime is ready. Agents do real work only after a Runtime is installed and signed in. Install one, sign in, then run linger-computer runtime check. This page compares the three Runtimes, shows how to check them, and explains what each check status means.

​
Which Runtime should I pick?

Pick the Runtime that you already use and pay for. The three differ in how they sign in and in how much Linger reviews their actions. The Guardrail checks each desktop, browser, and shell action before it runs. It applies rule-based checks. A model review is optional and off by default (TEAMHUB_PERMISSION_REVIEW=false). The review reduces mistakes, but it is not a sandbox. Pi runs without review, so give a Pi agent only work that you would let it do unchecked. See Approvals. Background work for memory and the personal memory runs as model calls on the Codex or Claude Code of your own computer. Those calls count against that account. If you use only Pi, also sign in to Codex or Claude Code on the same computer.

​
Install and sign in to a Runtime

Install at least one Runtime on the computer that runs your agents. Use the instructions from the Runtime’s own provider, then sign in. The desktop app’s setup can also do this step. It installs a Runtime if you have none and reuses tools and credentials that are already there. See Install the desktop app. Agents act with the access that the Runtime has on that computer: your files, signed-in apps, and shell. Connect only computers whose owner accepts that.

​
Check the Runtime

Run linger-computer runtime check to see which Runtimes are ready. The check works without a Linger sign-in and without a running Computer Daemon.
The command checks whether each Runtime is installed, signed in, and compatible. To check one Runtime only, name it: codex, claude, or pi.
If the check cannot find a Runtime, point it at the executable with --codex-path, --claude-path, or --pi-path.

​
Confirm that the model answers

Add --live to ask each ready Runtime for a model reply. Use it when the plain check passes but agents still do not answer.
A live check is a real model call, so it counts against your account. For scripts, add --json to print machine-readable results.

​
Read the check statuses

Each Runtime gets one of five statuses. The command exits with 0 when at least one checked Runtime is READY. A check of one Runtime exits with 13 for NOT_INSTALLED, 11 for AUTH_REQUIRED, 14 for INCOMPATIBLE, and 1 for ERROR.

​
Fix “No runtime is ready yet”

This message means that the computer is connected, but no Runtime on it is ready.
1

Run the check on that computer

2

Install or sign in

Fix the Runtime that the check names.
3

Confirm with a live check

Replace codex with claude or pi if you use another Runtime.
If the Runtime is ready but agents still do not answer, check the Computer Daemon. Run linger-computer daemon status and linger-computer doctor. Read linger-computer daemon logs, then try linger-computer daemon restart.

​
Keep learning

How Linger works

See how the Runtime fits with Computers and Rooms.

Local models

Point your Runtime at models on your own machine.

Approvals

Learn what the Guardrail checks before an agent acts.

Troubleshooting

Solve connection and agent problems.