docs(readme): document the chatgpt subscription provider

Author: cloudbridgeuy

README.md

@@ -9,6 +9,8 @@ llm-stream is a Rust library and CLI tool for streaming interactions with Large
 ## Features
 
 - Streaming support for various LLM providers
+- Sign in with a ChatGPT subscription — reach GPT models without a metered `OPENAI_API_KEY`
+- Reasoning-model support: effort levels and optional reasoning summaries
 - Easy-to-use API for integrating LLM capabilities into Rust applications
 - Command-line interface for quick interactions with LLMs
 
@@ -20,7 +22,7 @@ To use llm-stream in your Rust project, add the following to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-llm-stream = "0.1.0"
+llm-stream = "0.5.0"
 ```
 
 ### CLI Tool
@@ -86,6 +88,70 @@ For more CLI options and usage information, run:
 llm-stream --help
 ```
 
+### ChatGPT subscription
+
+`llm-stream` can reach GPT models through a ChatGPT subscription instead of a metered
+`OPENAI_API_KEY`. Sign in once, in a browser:
+
+```bash
+llm-stream --login          # opens a browser; prints: signed in as you@example.com (plus)
+llm-stream --login-status   # who is signed in, and how long the access token has left
+llm-stream --logout         # deletes the stored credentials
+```
+
+Then talk to it with `--api chatgpt` (aliases: `chat-gpt`, `codex`):
+
+```bash
+llm-stream --api chatgpt "say PONG"
+llm-stream --api chatgpt --model gpt-5.6-sol --reasoning-effort high "explain this diff"
+llm-stream --api chatgpt --reasoning-summary "what is 17 * 23?"
+```
+
+| Flag | What it does |
+| --- | --- |
+| `--login` | Browser sign-in (OAuth 2.0 + PKCE). Stores credentials at `~/.config/llm-stream/auth.json`, mode `0600`. |
+| `--login-status` | Prints the signed-in account, plan, and remaining access-token lifetime. Refreshes nothing. |
+| `--logout` | Deletes the stored credentials. Succeeds whether or not any existed. |
+| `--models` | Asks the server which models this account may use. Spends a little quota — see below. |
+| `--reasoning-effort <low\|medium\|high\|xhigh>` | How hard the model should think. Also settable per preset, or as `reasoning_effort` in `config.toml`. |
+| `--reasoning-summary` | Streams the model's reasoning summary to **stderr**, then a `---` rule, then the answer on stdout. The model decides whether to produce a summary at all. |
+
+This endpoint ignores `--temperature`, `--top-p`, and `--top-k`; use `--reasoning-effort`
+instead. It does not use `--api-key` or `--api-env`. Passing any of them prints a warning
+to stderr and continues.
+
+Answers always go to stdout and everything else to stderr, so `llm-stream --api chatgpt
+"..." | cat` gives you the answer and nothing else.
+
+#### Which models work
+
+The server decides, per plan and per rollout, so `llm-stream` ships **no local allowlist**:
+`--model` is passed through verbatim and the server's own refusal is what you see. Ask your
+own account:
+
+```bash
+llm-stream --models
+```
+
+It probes one model at a time and prints a table. Each accepted probe opens a real request
+and spends a small amount of your subscription's Codex allowance, so it warns you before it
+starts.
+
+Accepted on a Plus account on **2026-07-28**: `gpt-5.6-sol` (the default), `gpt-5.6-terra`,
+`gpt-5.6-luna`, `gpt-5.5`, `gpt-5.4`. Refused on the same account: `gpt-5.3`, `gpt-5.1`,
+`gpt-5-codex`. **That list will drift.** It is a snapshot of one account on one day, not a
+contract — and a model missing from it is still worth trying with `--model`.
+
+#### What this actually talks to
+
+This provider posts to `https://chatgpt.com/backend-api/codex/responses`, an **undocumented
+endpoint internal to OpenAI's Codex CLI**. It is not a public API and there is no
+specification to appeal to. OpenAI may change, gate, or block it at any time without notice,
+and using it from a client other than Codex is a gray area under OpenAI's terms of use.
+Requests draw on your ChatGPT subscription's Codex allowance, not on a metered API key.
+
+Use it knowing that.
+
 ## Contributing
 
 Contributions are welcome! Please feel free to submit a Pull Request.

crates/llm_stream/README.md

@@ -46,7 +46,27 @@ echo -n "What is the capital of" | llm-stream - "France?"
 
 ### Configuration ⚙️
 
-`llm-stream` uses a TOML configuration file to manage settings, API keys, and other customizations. The default configuration file is located at `~/.config/llm-stream.toml`.
+`llm-stream` uses a TOML configuration file to manage settings, API keys, and other customizations. The default configuration file is located at `~/.config/llm-stream/config.toml`. Run `llm-stream --config` to print the path in use, or `llm-stream --dir` for the directory.
+
+### ChatGPT subscription 🔐
+
+Reach GPT models through a ChatGPT subscription instead of an `OPENAI_API_KEY`:
+
+```bash
+llm-stream --login                                    # browser sign-in, once
+llm-stream --api chatgpt "say PONG"                   # stream an answer
+llm-stream --models                                   # which models this account may use
+llm-stream --api chatgpt --reasoning-effort high "explain this diff"
+```
+
+Credentials are stored at `~/.config/llm-stream/auth.json` with mode `0600`.
+`--login-status` shows who is signed in; `--logout` removes the credentials.
+
+This provider rides an **undocumented endpoint internal to OpenAI's Codex CLI**, which
+OpenAI may change or block without notice, and it spends your subscription's Codex
+allowance rather than a metered API key. See the
+[project README](https://github.com/cloudbridgeuy/llm-stream#chatgpt-subscription) for the
+full details, the verified model list, and the caveats.
 
 ### Presets
 

lib/llm_stream/README.md

@@ -7,11 +7,18 @@ This library provides a streamlined approach to interacting with Large Language
 
 ## Supported Providers
 
-- **OpenAI:** Access the powerful GPT models through OpenAI's API.
+- **OpenAI:** Access the GPT models through OpenAI's API.
 - **Anthropic:** Utilize Anthropic's Claude models for various language tasks.
 - **Google:** Integrate Google's Gemini family of models.
-- **Mistral:** Leverage Mistral's language models for advanced capabilities.
-- **GitHub Copilot:** Access code-generation capabilities powered by GitHub Copilot.
+- **Mistral:** Leverage Mistral's language models, including a fill-in-the-middle endpoint.
+- **DeepSeek:** Stream both the answer and the model's reasoning.
+- **Groq:** Low-latency inference over open models.
+- **Ollama:** Talk to models running locally.
+- **Jina:** Embeddings and reranking.
+- **ChatGPT (subscription):** Reach GPT models with a ChatGPT subscription's OAuth
+  credentials instead of a metered API key, over OpenAI's private Responses endpoint.
+  Undocumented and subject to change without notice — see the
+  [project README](https://github.com/cloudbridgeuy/llm-stream#chatgpt-subscription).
 
 ## Key Features
 
@@ -26,7 +33,7 @@ Add the following dependency to your `Cargo.toml` file:
 
 ```toml
 [dependencies]
-llm-stream = "0.1.3"
+llm-stream = "0.5.0"
 ```
 
 ## Usage