omgvibe: a one-command wizard for pointing four coding CLIs at OhMyGPT

What omgvibe actually does

omgvibe is a zero-install interactive wizard aimed at developers who use AI coding assistants but do not want to edit provider configuration by hand. It targets four CLIs: CodeX, Claude Code, OpenCode, and CodeWhale. The wizard connects each of these to OhMyGPT's unified /v1 API. The README positions it as a setup tool, not a runtime proxy. It does not run models itself. It installs the chosen CLI globally, probes three relay endpoints, writes config files, and creates backups before touching existing settings. The main audience is someone who already has an OhMyGPT API key and wants a guided path from key creation to a working CLI.

The endpoint selection mechanism

The most distinctive part of omgvibe is its endpoint probing. OhMyGPT exposes three /v1 endpoints: apic1.ohmycdn.com for Cloudflare Enterprise CDN outside mainland China and the US, api.ohmygpt.com for US direct, and cn2us02.opapi.win for China-optimized CDN. The wizard runs GET /v1/models on all three, five attempts each, then auto-selects the reachable endpoint with the lowest average latency. That is a concrete mechanism, and it matters because a relay that works in one region can be slow or blocked in another. The README does not say how it handles the case where all endpoints fail, or whether it retries after a timeout. The latency test is a real feature, but it is only as good as the network conditions at the moment you run it.

What it writes to your disk

Each target CLI gets a different config format, and omgvibe writes them all. For CodeX it writes ~/.codex/config.toml with a provider named omg, wire_api = "responses", a base URL of https://<endpoint>/v1, and experimental_bearer_token. It also writes ~/.codex/models.json with a full model catalog, defaulting to gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. For Claude Code it writes ~/.claude/settings.json with an env block that sets ANTHROPIC_MODEL and the Opus/Sonnet/Haiku slots to claude-opus-5, claude-sonnet-5, and claude-fable-5. For OpenCode it writes ~/.config/opencode/opencode.json with a custom omg provider using @ai-sdk/openai-compatible. For CodeWhale it writes ~/.codewhale/config.toml with provider = "openai" and a base_url, defaulting to alibaba:deepseek/deepseek-v4-flash-0731. The README also mentions an optional full reset for CodeX that moves the entire ~/.codex folder to a timestamped backup and rebuilds a fresh one, offered when you hit 400 errors or a missing model list.

Getting it running

The quick start is short. You need Node.js 18+ and npm. Then run npx omgvibe. The wizard asks you to choose a language (English, Simplified Chinese, or Japanese, with automatic locale detection), pick a CLI (CodeX is the default), confirm whether to run the official npm install -g for that CLI, paste your OhMyGPT API key, review latency results, pick default models, and confirm the write. Backups are created with names like config.toml.20250101-103000.bak next to the original files. For local development, the README lists npm install, npm run lint, npm run build, and running node dist/index.js after building. There is no mention of environment variables for the API key, so the paste step is the only path.

Where it can fail or be the wrong tool

The obvious limitation is that omgvibe only works with OhMyGPT's API. If you use a different provider or a self-hosted gateway, this wizard is useless. It also hard-codes a specific set of model names, like gpt-5.6-sol and claude-opus-5. If OhMyGPT updates its catalog or your plan does not include those models, the wizard may write configs that point at non-existent models. The README mentions a full reset for CodeX when you hit 400 errors, which suggests that model mismatches are a known failure mode. Another risk: the wizard runs npm install -g for the chosen CLI, which changes your global environment. If you already have a carefully tuned config, the backup mechanism helps, but you still need to verify the new files. The endpoint latency test is a snapshot; a fast endpoint at setup time can be slow later.

A real alternative: manual config editing

The alternative is to configure each CLI yourself. For CodeX, you would edit ~/.codex/config.toml directly, setting the provider, wire_api, base_url, and bearer token. For Claude Code, you would edit ~/.claude/settings.json to add the env block. For OpenCode, you would edit ~/.config/opencode/opencode.json. The difference is that manual editing gives you full control over every key and model name, and you do not need to trust a third-party wizard with your API key. The cost is that you must know the exact schema for each CLI, and you have to test endpoints yourself. omgvibe automates the probing and the backup, which is its main advantage. Manual editing is more transparent and avoids the possibility that the wizard writes something unexpected.

Maintenance, license, and trust considerations

The repository is under the MIT License, which means you can use, modify, and redistribute it freely, with no copyleft obligations. There are no recent releases listed, and the last push is unknown, so maintenance status is unclear. The README points to docs for each CLI's reference, which suggests the project tracks upstream config changes, but without a release history you cannot tell how quickly it adapts. The wizard writes API keys into plain text config files, which is standard for these CLIs but worth remembering. The backups are a good safety net, but they only help if you check them before the next run overwrites them. Since the project is a wrapper around OhMyGPT's API, its long-term usefulness depends on OhMyGPT's endpoint stability and model naming. If you use omgvibe, you are implicitly accepting that a third-party script will write to your home directory and run global npm installs.

Editorial conclusion

Adopt omgvibe if you already use OhMyGPT and want a reproducible way to wire CodeX, Claude Code, OpenCode, or CodeWhale to its relay without hand-editing config files. Skip it if you are not comfortable sending API keys through a third-party wizard or if you need to support CLIs outside the four listed. Before running it, verify that the endpoint list and model names match your OhMyGPT plan, and inspect the generated config files and .bak backups before deleting anything. The wizard's real test is whether the fastest endpoint it selects stays reachable in your network; if not, you can rerun it and pick a different endpoint manually.

Frequently asked questions

Official sources

1. Official README
2. Project repository
3. README
4. Releases