The Claude Code Changelog
All Episodes

Claude Code Goes Self-Hosted

We break down Anthropic’s new claude self hosted runner for Team and Enterprise users, showing how it lets Claude Code sessions run inside your own secure infrastructure instead of the cloud. The episode also covers the latest fail-fast permission checks, clearer spend-limit warnings, workspace trust prompts, and Remote Control improvements for attached images.

Show Notes


Chapter 1

Self Hosting Claude Code Execution with claude self hosted runner

Lachlan Reed

You know, James, if you work at a big corporate, you have probably felt this pain. You want to use the flashy web or mobile version of Claude Code from your couch, but your security team practically tackles you because your repository lives behind a strict firewalled VPC.

James Turner

Oh, absolutely. Corporate security hates external cloud agents touching internal code. But this episode is brought to you by Jellypod AI, and Anthropic actually just dropped a massive solution in version two point one point two hundred twenty four with the new command, claude self hosted runner.

Lachlan Reed

Yeah, it is brilliant mate! Basically, running claude self hosted runner with the base dir flag set to slash workspace slash project turns your own machines or containers into a place Claude Code web, mobile, and desktop sessions can run. So for Team and Enterprise accounts, the model can edit code and fire off shell commands right inside your secure infrastructure instead of Anthropic cloud boxes.

James Turner

That is huge for compliance! And they immediately followed up in version two point one point two hundred twenty five with a crucial fail fast fix. Previously, if your base dir path was invalid or unwriteable, the runner would register anyway and then silently fail on every single session, which drove developers crazy.

Lachlan Reed

Oh man, silent failures give me total flashbacks to my early dev days! I once pushed a bad background update at midnight that silently tanked a staging environment. Now, the runner does a strict permission check right on startup and crashes with an explicit error before it takes any jobs. I actually spun up a sandboxed runner container on my backyard shed server yesterday, and knowing it won't ghost me if a volume fails to mount is massive peace of mind.

James Turner

Right on. And version two point one point two hundred twenty five packed in some great quick hit updates too. Gateway spend limit warnings now break down exact dollar caps, reset times, and custom admin messages. Plus, running claude agents now triggers workspace trust prompts in untrusted directories, and Remote Control handles attached images directly without wasting extra disk tool calls.

Lachlan Reed

Keeps things secure and snappy. Good chatting, James!