Skip to content
OhMyGPT

Search

Search pages and models…

Use 219 AI models from Claude Code and Codex

omgvibe is an open-source command-line tool. Run ov claude and Claude Code opens already connected to OhMyGPT. 85 text models cost less here than at the official API. All models

Source on GitHub
npm install -g omgvibe
ov claude

Needs Node.js 20.12 or newer. ov claude leaves Claude Code’s config files alone, so running claude on its own works as before.

Supported agents

omgvibe starts the agent already installed on your computer. If it isn’t there, omgvibe offers to install it.
  • Claude Codeov claude
  • Codexov codex
  • opencodeov opencode
  • Kilo Codeov kilo
  • Qwen Codeov qwen
  • Piov pi
  • CodeWhaleov codewhale

Commands

Run ov on its own for a menu with all of these.
ov <agent> [args…]
Start the agent connected to OhMyGPT. Anything after the agent name goes straight to the agent, as in ov claude -p "review this diff".
ov <agent> --model <id>
Start on another model, such as Claude Code on GLM. If you don’t remember the model ID, search with --pick.
ov models [search…]
List models with their context window and price. Add words to search.
ov setup [agent…]
Save the OhMyGPT address and key in the agent’s config files, so typing claude directly or using an IDE extension works too.
ov reset [agent…]
Undo what setup changed.
ov endpoint
Measure the regional endpoints again and pick one. One of them is tuned for mainland China.
ov doctor
Run this first when something won’t connect. It checks your key and endpoint, then tries Chat Completions, Responses and Messages once each on the free test model.
ov login
Save or replace your API key.
npx omgvibe <agent>
Use it without installing, for example in CI or a container.

First run

The first run asks you to paste your API key once and finds the fastest endpoint from where you are. After that, the agent starts straight away.
$ ov claude│◇  Sign in to OhMyGPT│◇  Endpoints measured││  Cloudflare global acceleration       https://apic1.ohmycdn.com          113 ms│  US direct                            https://api.ohmygpt.com             96 ms  fastest│  Asia optimized                       https://c-z0-api-01.hash070.com    141 ms│  Asia (optimized for mainland China)  https://cn2us02.opapi.win          132 ms│  Singapore                            https://api-ep-sg1.ohmygpt.com     167 ms│  Europe (Germany)                     https://api-ep-eu1.ohmygpt.com     188 ms│◇  Open the API keys page in your browser to create or copy a key?│  Yes│◇  Paste your API key│  ▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪│◇  Key accepted: account #1024, 1,000 requests per minute│◆  Key saved to ~/.omgvibe/config.jsonClaude Code on OhMyGPT: the agent's default model via US direct
$ ov codex --model zhipu/glm-5.3Codex on OhMyGPT: zhipu/glm-5.3 via US direct

Why OhMyGPT

TEE proof of which provider answered
With TEE proof on for a key, responses carry a signed receipt you can verify, recording the provider host (such as api.anthropic.com) and what it returned. Learn more
Automatic failover
Each model has several providers. If one errors or is rate limited, the request moves to the next before any output starts. Provider uptime is public. Learn more
Billed on actual usage
You pay for the usage the provider reports, with cache priced separately. Failed requests are free, and every request has an itemized receipt. Learn more
Native APIs, no code changes
Native OpenAI Chat Completions and Responses, Anthropic Messages and Gemini APIs. Claude Code and Codex connect by changing the base URL. Learn more
Your providers, your order
Pick which providers to use per request, set their order, or exclude one. If you don't choose, OhMyGPT picks for you and fails over to the next. Learn more
Payment and invoices
Pay by card, Alipay, WeChat Pay and more. Online payment comes with a downloadable invoice and receipt.