WindowsForum Diagnostics 2.5.2: Use Your ChatGPT or Claude Subscription — No API Key Required

[ChatGPT]

WindowsForum Diagnostics 2.5.2 is now rolling out through the Microsoft Store.
This is a focused update with one major new capability: the AI assistant can now use the AI subscription you already pay for. If you have ChatGPT Plus, Pro, or Team or Claude Pro or Max, you can sign in once and use the assistant without creating an API account, purchasing API credits, or pasting an API key.

---

Highlights​

• Sign in with ChatGPT. A new provider routes requests through OpenAI's Codex CLI running locally on your PC. Usage is billed to your existing ChatGPT subscription—no OpenAI API key required.
• Sign in with Claude. A second provider integrates with Anthropic's Claude Code CLI over the Agent Client Protocol (ACP), the same integration approach used by Microsoft's Intelligent Terminal. Responses stream live as they're generated.
• Model pickers everywhere. Every provider in Settings now includes a model selector. Cloud providers fetch their available models directly from the service, local providers report installed models, and subscription providers expose a curated catalog.
• Smarter Auto routing. Auto mode remains local-first, but if you're signed into ChatGPT or Claude, those subscriptions now take priority over metered API keys so you don't accidentally spend pay-per-token credits while a flat-rate subscription is available.

---

How the Subscription Providers Work​

The guiding principle behind this feature is simple:
WindowsForum Diagnostics never accesses your account credentials.
The application does not implement OAuth, store authentication tokens, or read any credential files maintained by OpenAI or Anthropic.
Instead, it detects the official CLI you already have installed—either OpenAI Codex CLI or Claude Code CLI—and lets that software handle authentication entirely.
When you click Sign in with ChatGPT or Sign in with Claude, the vendor's own authentication flow opens in your browser exactly as if you had launched the CLI manually from a terminal.
The application only knows one thing:

• Signed in
• Not signed in

Nothing more.

---

Built With Security in Mind​

Requests follow the same philosophy.
Prompts are passed directly to the official vendor tooling over standard input—not through shell command lines—and every execution is intentionally restricted.

ChatGPT (Codex CLI)​

• Runs inside a read-only sandbox
• Uses an empty working directory
• Cannot read your files
• Cannot modify your system

Claude Code​

• Runs through Microsoft's Agent Client Protocol (ACP)
• All tool permissions are explicitly denied
• Claude responds only to the diagnostic context supplied by the application

Additionally, WindowsForum Diagnostics strips API-key environment variables before launching either provider. If your system happens to contain an OPENAI_API_KEY or ANTHROPIC_API_KEY, those variables cannot silently switch billing from your subscription to pay-per-token API usage.

---

Compliance​

OpenAI has publicly supported third-party applications using ChatGPT subscriptions through Codex.
The Claude integration mirrors Microsoft's own Intelligent Terminal implementation by using the official Claude Code CLI directly under your own authenticated account.
No credentials are extracted.
No authentication is spoofed.
No requests are proxied through WindowsForum servers.
Everything happens using the vendor's officially supported tooling.

---

Getting Started​

Provider	Requirements	Setup
ChatGPT via Codex CLI	ChatGPT Plus, Pro, or Team + npm install -g @openai/codex	Settings → Provider Setup → ChatGPT via Codex CLI → Sign in with ChatGPT
Claude via Claude Code	Claude Pro or Max + npm install -g @anthropic-ai/claude-code	Settings → Provider Setup → Claude via Claude Code CLI → Sign in with Claude

The application automatically detects both CLIs, whether installed through npm or native installers.
If one isn't found, you'll see installation guidance along with the option to specify a custom executable path if your tools are installed in a non-standard location.
Claude downloads a small ACP adapter the first time it's used, so the initial request may take a few extra moments. After that, responses stream token-by-token.

---

Model Selection Everywhere​

Previous versions relied on fixed or manually entered model names for many providers.
Version 2.5.2 introduces model selection throughout the application.

Cloud Providers​

• OpenAI
• Anthropic
• Gemini
• DeepSeek

As soon as an API key is entered—even before saving—the application retrieves your available chat-capable models directly from the provider.

Custom OpenAI-Compatible Endpoints​

Providers such as:

• OpenRouter
• Groq
• Self-hosted proxies

automatically populate the model selector using the endpoint's /v1/models response.

Local Providers​

Both Foundry Local and Ollama now display exactly the models available on your local server.

Subscription Providers​

ChatGPT via Codex and Claude Code include a curated list of current models, with CLI Default remaining the recommended choice if you simply want the vendor's default configuration.
OpenAI and Foundry Local also gain model override support for the first time, so you're no longer limited to the application's default selections.

---

Availability​

Version 2.5.2 has been submitted to the Microsoft Store and will begin rolling out automatically once Microsoft certification is complete.
The Microsoft Store release remains the recommended installation because it is Microsoft-signed and is currently the only build that supports Phi Silica acceleration on compatible Copilot+ PCs.
Microsoft Store
WindowsForum Diagnostics - Free download and install on Windows | Microsoft Store
Portable builds and release notes
Releases · faratech/wfdiag · GitHub

---

As always, WindowsForum Diagnostics remains read-only by design.
The assistant can inspect your system, explain issues, and recommend solutions, but it will never make repairs without your explicit confirmation—and that philosophy isn't changing.
Questions, feedback, feature requests, and bug reports are all welcome in this thread.