Connect my computer: works from Docker and on Windows

- Docker passes BROKE_HOSTED through (setting it in .env did nothing before)
- the helper finds claude.exe / claude.cmd on Windows (PATH uses ';' there),
  runs a .cmd through cmd.exe with the question on stdin only, and keeps the
  Windows variables Claude needs to find its login; opencode on Windows only as .exe
- the page gives Windows a PowerShell command (curl.exe, file then node)
- Copy works on a plain http LAN address

Tested on Windows end to end with a stand-in claude.cmd: pairing, a question with
quotes, & and %, and every lockdown flag arriving intact.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
MangaValkandClaude Opus 5.5 committed 2026-11-17 20:50:41 +01:00
1 parent ea53718e5d
commit 0232766260
4 files changed
+54 -12

No files matched your search

+10
View File
@@ -157,6 +157,16 @@ Read what changed (`git show <commit>`). Take everything with `git merge upstrea
Things to check in upstream changes: new outside calls (`fetch(` to a new host), new environment variables, and changes to `scripts/start.mjs` or `src/lib/llm/`.
## Use Claude from your own computer
brokeless runs in Docker, so it can't use a Claude login of its own. Instead, Claude Code keeps running on your PC and brokeless sends it questions:
1. On your PC, install Claude Code (`npm install -g @anthropic-ai/claude-code`, or Anthropic's installer) and run `claude` once to sign in.
2. Add `BROKE_HOSTED=1` to `docker/.env` and run Compose Up again.
3. In brokeless: Connect your AI, **Connect my computer**. Run the command it shows on your PC (in PowerShell on Windows) and leave it open. Then pick Claude.
Claude only answers while that command runs. It starts Claude with every tool off, in an empty folder, so it can't touch your files. Works on Windows, macOS and Linux.
## How AI stays private
Every AI path goes through one gate. A cloud model gets a summary: rounded amounts, names replaced by tokens. Full detail goes only to a local model, which is what Ollama is here. You can turn AI off in Settings. Each answer runs with no way to change your data.