R Assistant Gateway

Love Codex models, hate API prices? ☀️ 🌍 🌙 Use your existing ChatGPT/Codex subscription sign-in with Posit Assistant in RStudio or Positron.

r-assistant-gateway is a small local compatibility bridge between Posit Assistant's OpenAI Responses client and ChatGPT/Codex. It uses @carl-stone/openai-oauth to sign in with your ChatGPT account and translates Posit Assistant's requests into the request contract accepted by ChatGPT/Codex.

It is intentionally only that bridge. It is not a general OpenAI proxy or an API-key replacement for other applications.

Unofficial community project. This project is not affiliated with, endorsed by, or supported by Posit or OpenAI.

Before you start

You need:

• Node.js 24 or newer;
• RStudio or Positron with Posit Assistant installed; and
• a ChatGPT/Codex Plus or Pro subscription.

Tested with Posit Assistant 1.3.0 and 1.3.1. Compatibility with older versions is unknown.

The gateway uses your ChatGPT/Codex sign-in. You do not need an OpenAI API key.

Quick start

With Node.js installed, run these commands in a Terminal window:

npm install --global r-assistant-gateway
r-assistant-gateway login
r-assistant-gateway

The last command keeps the gateway running in the foreground. Leave that Terminal window open while you use Posit Assistant.

To run it in the background instead, use:

r-assistant-gateway --detach

Connect Posit Assistant

Choose the OpenAI provider in either IDE:

• RStudio: in the Posit Assistant pane, select gear > Configure AI providers > OpenAI.
• Positron: open the Command Palette, run Authentication: Configure Language Model Providers, and select OpenAI. The separately named OpenAI Compatible provider defaults to Chat Completions; this gateway uses Responses.

Enter the same connection settings in either IDE:

| Setting | Value | | --- | --- | | Base URL | http://127.0.0.1:10532/v1 | | API key | local-gateway |

local-gateway is only a non-secret placeholder required by the OpenAI setup form. The gateway does not use it as an OpenAI API key; upstream requests use your ChatGPT/Codex sign-in. A non-empty value also lets Posit Assistant discover the models available through your account from the gateway. One running gateway can serve both IDEs on the same computer. In Positron, open the chat with View: Show Posit Assistant. See the official Positron provider instructions for the provider dialog.

A model is missing from the Positron selector

Posit Assistant 1.3.1 in Positron filters OpenAI model discovery to IDs beginning with gpt-5, gpt-4, or o. This hides gpt-6-astra even when the gateway correctly lists it. Add the model explicitly using Open AI Provider Settings (JSON), which opens ~/.posit/ai/providers.json.

Merge this example into your existing providers.openai settings, preserving any other providers or custom models:

{
  "providers": {
    "openai": {
      "baseUrl": "http://127.0.0.1:10532/v1",
      "models": {
        "custom": [
          {
            "id": "gpt-6-astra",
            "name": "GPT-6 Astra",
            "protocol": "openai-responses",
            "maxContextLength": 272000,
            "supportsTools": true,
            "supportsImages": true,
            "supportsToolResultImages": true,
            "supportsWebSearch": false,
            "thinkingEffortLevels": ["low", "medium", "high", "xhigh", "max", "ultra"]
          }
        ]
      }
    }
  }
}

Reopen the model selector; if needed, run Developer: Reload Window. This adds Astra alongside automatically discovered models. It does not grant access to models unavailable to your account. The example reflects the Astra metadata used in the September 2026 validation; check current model capabilities when adding other models. See Posit’s custom model settings reference.

Start, check, and stop the gateway

These commands are useful when the gateway runs in the background:

r-assistant-gateway status
r-assistant-gateway logs --follow
r-assistant-gateway doctor
r-assistant-gateway stop

Run r-assistant-gateway again to start it in the foreground after stopping it. The default port is 10532, which matches the Posit Assistant URL above. You can choose another port with, for example, --port 10533; if you do, change the provider's base URL to match.

What the gateway translates

Posit Assistant uses the OpenAI Responses wire format in both IDEs when the OpenAI provider is selected. For current Codex models it can send developer and user input, images and files, tool definitions, function calls, structured function outputs, encrypted reasoning, and explicit prompt-cache controls.

ChatGPT/Codex accepts most of that request directly. The gateway removes only known incompatible fields:

• prompt_cache_options;
• prompt_cache_breakpoint markers on Posit's input content parts;
• previous_response_id and legacy prompt-cache retention fields; and
• any other root field outside the current Codex request contract, including Posit's max_output_tokens.

The model, messages, images, files, reasoning, tools, function calls and structured results, cache key, and streaming response remain intact. The adapter never mutates the request object supplied by its caller.

The OAuth runtime handles sign-in, token refresh, model discovery, upstream transport, and server lifecycle. The gateway defaults its optional Responses history to process-local memory so ID-based continuations can be resolved when needed. That history is discarded whenever the gateway stops.

Tested versions

Gateway 0.2.x has been tested with Posit Assistant 1.3.0 and 1.3.1. Compatibility with older versions is unknown. These are tested configurations, not minimum version requirements:

| IDE | Tested Posit Assistant version | | --- | --- | | RStudio | 1.3.0 | | Positron | 1.3.1 |

Troubleshooting

Run the read-only diagnostic report with:

r-assistant-gateway doctor

It reports the installed gateway and OAuth runtime versions, Posit Assistant in RStudio (positAssistant) and Positron (positronAssistant), and the active gateway’s health. Each Assistant entry includes its version and installation path. The compatibility check recognizes the tested configurations: RStudio Assistant 1.3.0 or Positron Assistant 1.3.1, alongside the expected OAuth runtime. An unrecognized Assistant version is untested, not necessarily incompatible. An older or missing installation in the other IDE does not prevent success. The command still exits unsuccessfully if the gateway health check fails.

Positron detection checks ~/.positron/extensions, using the extension registry when available and ignoring removed extensions. For a custom extension directory (such as a --extensions-dir setup), set POSITRON_EXTENSIONS_DIR. Detection reports installed packages; it does not check whether an extension is enabled in the active Positron profile. The diagnostic report does not read conversations or credentials.

Common fixes:

• RStudio or Positron cannot connect: make sure the gateway is running and the base URL is exactly http://127.0.0.1:10532/v1.
• The port is busy: stop the other process, or start with r-assistant-gateway --port <number> and update the base URL.
• Sign-in fails: run r-assistant-gateway login again, then restart the gateway.
• A conversation fails after restarting the gateway: start a new Posit Assistant conversation. Temporary continuation state is cleared on restart.
• doctor does not recognize your Assistant version: versions 1.3.0 and 1.3.1 are tested; compatibility with older versions is unknown. A failed version check does not establish that your installation cannot work. For an administrator-managed or otherwise nonstandard installation, set POSIT_ASSISTANT_ROOT to RStudio’s pai/bin directory, or POSITRON_EXTENSIONS_DIR to Positron’s extension directory, before running doctor.
• doctor reports an unexpected OAuth runtime: reinstall the matching r-assistant-gateway release rather than upgrading its runtime directly.
• The background gateway is not working: run status, inspect logs, then use stop before starting it again.

Privacy and network behavior

By default, the gateway listens only on your computer at 127.0.0.1. It uses the same host behavior as openai-oauth. An explicit non-loopback --host can make the unauthenticated endpoint reachable by other computers; every client that can reach it can use your ChatGPT/Codex subscription. Keep the default unless that access is deliberate and protected by your network.

OAuth credentials and upstream transport are handled by the published OAuth runtime. Recent Responses items are held only in the running process, with default limits of 256 response-history entries and 2,000 items. They are never persisted by the gateway and are discarded when it stops or restarts.

On POSIX systems, gateway startup and login restrict the selected OAuth credential file to its owner (0600).

Diagnostics are off by default. If enabled with --diagnostics, they contain metadata only: request ID, model, schema-only removed-field patterns, cache breakpoint count, status, duration, and safely available token counts. Prompts, conversation content, tool arguments and results, credentials, headers, auth material, and reasoning content are not logged.

Advanced CLI options

The gateway accepts the same public commands and options as its OAuth runtime, including login, --host, --port, --models, --codex-version, --base-url, OAuth overrides, --no-open, login timeout, --detach, status, logs, stop, --responses-state, --responses-max-responses, and --responses-max-items.

There are two intentional Posit defaults: port 10532 and process-local Responses memory. You can override the port. Setting --responses-state stateless disables continuation by response or item IDs, so doctor does not consider that configuration healthy for Posit Assistant. doctor and --diagnostics are specific to this gateway.

OAuth runtime limitations

• Use an IPv4 host such as the default 127.0.0.1. The pinned OAuth runtime does not correctly advertise an IPv6 host.
• Cancelling a request in the client may not stop the upstream generation. The pinned runtime does not yet propagate HTTP disconnects to its upstream fetch.
• The delegated runtime also exposes unadapted chat-completion and image routes. They are outside this gateway's supported surface; this is another reason to keep the listener on the default loopback address.

Development

Start with the interface and reverse-engineering notes to see which traffic belongs to each IDE, Posit Assistant, the gateway, and Codex.

npm install
npm run verify
npm run check:contract
npm pack --dry-run

npm run verify performs typechecking, linting, unit tests, a TypeScript build, and a detached end-to-end CLI test using a Posit Assistant 1.3.0-shaped request. CI also performs an npm package dry run. A separate scheduled/manual workflow verifies that every root field forwarded by the adapter remains accepted by Codex. Dependabot watches npm and GitHub Actions dependencies.

Credits and license

This project uses @carl-stone/openai-oauth, a published fork of Evan Zhou's openai-oauth. The dependency ships its own Apache-2.0 license and notice; this repository's required attribution is included in NOTICE.

The original gateway code in this repository is copyright Carl Stone and is licensed under Apache-2.0. See LICENSE.

Legal

R Assistant Gateway is an unofficial community project. It is not affiliated with, endorsed by, or supported by Posit Software, PBC.

Posit, RStudio, Positron, and Posit Assistant are trademarks of Posit Software, PBC, all rights reserved, and may be registered in the United States Patent and Trademark Office and in other countries.