RD-X-96

RD-X-96 is a standalone coding assistant for your terminal. Connect your own API-key or subscription provider and work with your chosen model.

Fork source: https://github.com/EstarinAzx/codebuff-modded

Installation

npm install -g codebuff-mod

(Use sudo if you get a permission error.)

Quick start

cd ~/my-project
rdx           # aliases: cbm, codebuff-mod

Inside the CLI:

/providers:add <preset> <profile-name> <apiKey>

Presets:

| Preset | Default model | Notes | |---|---|---| | openai | gpt-5.1 | api.openai.com | | anthropic | claude-sonnet-4.5 | api.anthropic.com | | openrouter | anthropic/claude-sonnet-4.5 | openrouter.ai | | opencode | minimax-m2.7 | opencode.ai/zen/v1 | | opencode-go | glm-5 | opencode.ai/zen/go/v1 | | deepseek | deepseek-chat | api.deepseek.com | | gemini | gemini-2.5-pro | generativelanguage.googleapis.com | | mistral | mistral-large-latest | api.mistral.ai | | together | meta-llama/Llama-3.3-70B-Instruct-Turbo | api.together.xyz | | groq | llama-3.3-70b-versatile | api.groq.com | | grok | grok-4.6 | SuperGrok subscription; /providers:add grok [name], no API key | | codex | Account model catalog | ChatGPT subscription; /providers:add codex [name], no API key | | custom-openai | (yours) | Any OpenAI-compatible endpoint — needs <baseUrl> arg |

Then run any coding task. RD-X-96 uses the model from your active profile and sends requests directly to your provider.

Web search, browser and desktop control

A saved Codex profile supplies web search and source URLs without another search-service API key. Your main model can remain Grok or another provider. Search uses the Codex plan's allowance; the active Codex profile is preferred, otherwise the first configured Codex profile is used.

/browser on enables local browser control. It requires Node.js 22.15+ and Microsoft Edge on Windows or Google Chrome on Linux. /computer on enables Windows desktop control and requires uv; Python 3.13 and the pinned server are downloaded when needed. Both features start disabled and need no service API key. First setup needs internet.

Use /browser or /computer for status, and /browser off or /computer off to interrupt the active run and close that connector. Enabled tools act with your user access without per-action confirmation. Managed browser and desktop tools are excluded from PLAN mode. Existing provider profiles and user-defined MCP configuration are preserved.

This release ships Windows x64, Linux x64 and Linux arm64 binaries. See tool documentation for supported actions, dependency versions and configuration.

Grok subscription login

Run /providers:add grok (or /providers:add grok Work for a named account). Open the displayed xAI verification link, enter the code and approve. The CLI activates the profile after authorization succeeds; a denied or expired login leaves your current provider selected.

Grok uses xAI's subscription proxy and the Responses API. Run /model for live account model discovery, /model <id> to switch, or /providers:refresh-models to refresh the five-minute cache. Offline lists are labeled. Tokens refresh automatically and are stored separately per profile in ~/.config/manicode/grok-oauth.json; removing a Grok profile removes its stored tokens. /providers:test checks Grok authentication and catalog access without generating a paid request.

The integration follows the device-code protocol inspected in pi-grok. It does not change your xAI account's privacy settings.

Commands

| Command | What it does | |---|---| | /providers | List your profiles (* marks active) | | /providers:add <preset> <name> <apiKey> | Add a new profile, set active | | /providers:select <id\|name> | Switch active profile | | /providers:remove <id\|name> | Remove a profile | | /providers:test | Check the active provider (Grok: authentication/catalog; API-key providers: 1-token ping) | | /providers:refresh-models | Clear the active profile's models cache (OAuth: 5 minutes; other providers: 24 hours) | | /model | Show current model + live-probe available ids | | /model <id> | Swap model on the active profile | | /browser [on\|off\|status] | Enable, stop or inspect local browser control | | /computer [on\|off\|status] | Enable, stop or inspect Windows desktop control | | /mode:default /mode:lite /mode:max /mode:plan | Switch agent mode (mod-* templates in .agents/) |

Your profiles live at ~/.config/manicode/providers.json (chmod 0600). API keys are masked in all log output.

Knowledge files

Add a knowledge.md anywhere in your project to give the agent persistent context. The agent reads + writes them as it works.

Troubleshooting

Permission errors during install

sudo npm install -g codebuff-mod

If still broken, reinstall Node.

Binary download fails

The launcher fetches the platform binary from GitHub Releases of EstarinAzx/codebuff-modded on first run. If you're behind a proxy, set HTTPS_PROXY:

export HTTPS_PROXY=http://your-proxy-server:port   # bash/zsh
$env:HTTPS_PROXY = "http://your-proxy-server:port" # PowerShell
set HTTPS_PROXY=http://your-proxy-server:port      # CMD

Also supported: HTTP_PROXY, NO_PROXY (with comma-separated hostnames). URL-embedded credentials work (http://user:pw@proxy:port).

"No active BYOK profile and no Codebuff backend configured"

You haven't added a provider profile yet. Run /providers:add (see above).

Override binary download URL (testing)

export CODEBUFF_MOD_RELEASE_URL=https://example.com/codebuff-mod-linux-x64.tar.gz

Use the legacy Codebuff backend instead (advanced)

export CODEBUFF_USE_BACKEND=1
export NEXT_PUBLIC_CODEBUFF_APP_URL=https://codebuff.com

This restores the upstream behavior (requires a real codebuff.com account + API key).

License

MIT. Built on top of Codebuff (Apache-2.0).

Issues

https://github.com/EstarinAzx/codebuff-modded/issues