Login flow is basically unusable on Windows — please make authentication work before release

Login flow is basically unusable on Windows — please make authentication work before release

I’m honestly very frustrated with the current login experience on Windows.

I downloaded the Windows executable and tried to use jcode natively on Windows, but the authentication flow is almost unusable.

Problems I hit

1. jcode login --provider openai-compatible asks for the API base URL.

2. After entering https://api.deepseek.com, it either skips the API key input directly or exits with:

   Error: No API key provided.
   

3. Inside the TUI, /login opens a provider selector, but mouse selection does not work, and the keyboard interaction is very unclear.

4. When selecting OpenAI-compatible, the UI shows opencode documentation:

   https://opencode.ai/docs/providers#custom-providers
   

   This is confusing because I am using jcode, not opencode.

5. The UI keeps showing:

   No models are available right now.
   

6. The client gets stuck in repeated reconnecting state:

   reconnecting (...)
   

7. The logs also show:

   Found existing OpenAI/Codex credentials from Codex at C:\Users\...\.codex\auth.json
   but jcode will not read them without confirmation.
   

   But there is no clear, working path to confirm this in the current UI.

User experience

From a user perspective, this is extremely frustrating. Before adding more agents, providers, UI features, or advanced workflows, the basic login flow really needs to be solid.

Right now, the first experience is:

• Download executable
• Launch jcode
• Try to login
• Cannot input API key reliably
• Cannot select provider clearly
• See opencode docs inside jcode
• Get “No models available”
• Get stuck reconnecting
• Check logs manually
• Still no obvious way to fix it

This is not a good onboarding experience.

What I expect

Please consider fixing the login/authentication flow before presenting this as ready for Windows users.

At minimum, I think jcode should provide:

• A reliable non-TUI login command for API key providers

• Clear documentation for where openai-compatible.env should be placed on Windows

• A command like:

  jcode login --provider openai-compatible --api-base https://api.deepseek.com --model deepseek-v4-flash
  

  Then securely prompt for the API key.

• Clear error messages when no model is available

• A way to skip reading Codex credentials if the user only wants OpenAI-compatible

• Better Windows terminal compatibility

• Avoid showing opencode docs without explanation, or clearly explain that jcode reuses opencode provider configuration

Final note

I really want to try this project, but login is the first step. If login does not work, nothing else matters.

Please make authentication reliable first.

Start by reproducing jcode login --provider openai-compatible on Windows, then trace the /login TUI flow and the openai-compatible.env and .codex/auth.json paths mentioned in the report. Done means API-key entry, provider selection, model discovery, reconnect behavior, and credential confirmation have clear working paths for Windows users.