Adom Hydrogen
Public Made by Adomby adom
The whole Adom experience as one signed native app: your editor, your AI agent, and the full electronics design toolchain, running locally on your own machine.
| Name | Last updated |
|---|---|
| nat-vs-mirrored-poster.jpg | 9d ago |
| nat-vs-mirrored.chapters.json | 8d ago |
| nat-vs-mirrored.mp4 | 9d ago |
| nat-vs-mirrored.srt | 8d ago |
| nat-vs-mirrored.video.json | 8d ago |
| nat-vs-mirrored.vtt | 8d ago |
| README.md | 8d ago |
NAT vs mirrored networking
Four minutes, narrated. Claude Code, Codex, Antigravity and Kimi Code each sign out and back in on mirrored networking, then again on NAT. An app dashboard runs on NAT, and you see the setup step that switches modes on its own. Recorded on a test PC running Hydrogen 1.0.701. The waiting is sped up (the corner shows how much), and email addresses are blurred. Download the video (captions: VTT, SRT).
Hydrogen runs your Linux workspace in WSL on your own Windows PC. WSL can connect that workspace to Windows in two ways, and which one works depends on the PC. This page covers both modes, the install problem the difference caused, and how Hydrogen handles each mode now.
The two modes
| Mirrored (the default) | NAT | |
|---|---|---|
| How it works | Linux shares Windows' network. localhost on Windows and localhost in Linux are the same place. |
Linux gets its own small network behind Windows. WSL forwards Windows' localhost into it. |
| AI sign-ins | The browser on Windows calls the AI's listener in the workspace directly. | Same, through WSL's localhost forwarding. Hydrogen checks the trip before relying on it. |
| App dashboards | Open at the same address in both modes (below). | Same. |
| Where it breaks | Some PCs block it: a security tool or a firewall policy stops Windows from reaching the workspace. | Turning off localhostForwarding in .wslconfig breaks it. Hydrogen turns it back on. A Windows program can also already hold an app's port. |
Why this became an install problem
On the PCs where mirrored is blocked, setup used to stop with a message that pointed the wrong way:
- The editor never loaded at "Sign in to Adom, then activate the editor", and setup halted there (#145).
- Windows could not reach a workspace that was running fine, and setup ended on "Couldn't build the workspace image". That sent people off to delete a good workspace (#157).
Both people who reported it had a working workspace that Windows could not reach. Install data showed the same pattern: setup halted at the editor step or the workspace step, and those people never got to use Hydrogen.
How Hydrogen handles it now
During install. The editor step signs you in to Adom and then checks that the editor really loaded. If it does not load on mirrored, or Windows cannot reach the workspace, setup:
- switches this PC to NAT once, recording your original setting so uninstalling puts it back,
- restarts WSL (unless another Linux distribution is running, in which case it stops and names it), and
- carries on.
A second failure in the same run stops and says why, so it never loops. Turning IPv6 back on from the setup card also puts you back on mirrored. (1.0.682, 1.0.686, 1.0.689)
Sign-ins on NAT. Hydrogen used to warn on every NAT workspace that AI sign-ins could not work. That was wrong, because WSL forwards Windows' localhost into the workspace. Now Hydrogen tests the real trip: it opens a one-off listener in the workspace and fetches it from Windows. It warns only when that fails, and then points you to signing in with an API key. When you click Sign in, Your AIs also checks that the AI's own callback port answers from Windows. If .wslconfig has localhostForwarding turned off, setup turns it back on and remembers your value. (1.0.671, 1.0.672)
How each AI signs in:
| AI | How sign-in finishes | NAT |
|---|---|---|
| Claude Code | Browser calls back to a localhost port in the workspace | Works through forwarding, port checked first |
| Codex | Browser calls back to localhost:1455 | Works through forwarding, port checked first |
| Antigravity | Google sign-in, handed back to the workspace | Works |
| Kimi Code | Device code: you approve in the browser, nothing calls back | Works, nothing to forward |
App dashboards. Hydrogen opens dock apps and apps your AI builds through its own proxy at an address like http://8893.localhost:47083/, or straight at the app's 127.0.0.1 port. Either way the address is the same in both modes, so nothing changes when you switch. On NAT, a program on Windows can already hold the port an app wants. The webview tab now names that program and asks your AI to use another port, instead of quietly showing the other program's page. (1.0.695)
Your choice, any time. Settings, Workspace networking switches between mirrored and NAT. WSL restarts and the editor reconnects on its own. (1.0.675)
The balance
- Mirrored by default: it is the simplest, and everything connects directly.
- NAT automatically, only on the PCs where mirrored is blocked, with nothing for you to diagnose.
- One setting to switch either way, which Hydrogen puts back if you uninstall.
Known rough edges (from making this video)
- The editor can take up to a minute to answer after a switch. Your AIs says "Your workspace is still starting" in the meantime. Once, after switching to NAT, it stayed stuck until the editor was reloaded. That is on the list for the next build.
- Windows may ask a firewall question for Adom Hydrogen the first time a sign-in runs on NAT. In the recording, the sign-in finished after pressing Cancel, and Windows did not ask again. Stopping Hydrogen from triggering the prompt at all is on the list.
Each behaviour above has a check in Hydrogen's promotion gate (mirrored-breaks-editor-nat, mirrored-unreachable-workspace-nat, wsl-nat-fallback, nat-signin-honest and nat-ai-app-webview), and every build is tested against it before it goes public.