Providers
Connect 95+ AI providers to kRouter: OAuth subscriptions, free services, free-tier and pay-per-token API keys, and browser-cookie logins.
A provider is an upstream AI service that kRouter routes to. kRouter supports 95+ of them, and you can add any number of your own OpenAI-compatible, Anthropic-compatible or embedding endpoints on top. The dashboard groups providers by how you connect them.
OAuth subscriptions
Sign in once with an account you already have. kRouter keeps the tokens refreshed and, where the provider reports it, tracks each account's quota.
- Claude Code (
cc): Claude Opus 4.8, 4.7 and 4.6, Sonnet 4.6, Haiku 4.5 - OpenAI Codex (
cx): GPT-5.5, GPT-5.4, GPT-5.3 Codex, plus image models that need ChatGPT Plus or higher - GitHub Copilot (
gh): GPT, Claude, Gemini and Grok models - Antigravity (
ag): Gemini 3.x, Claude Sonnet 4.6, Claude Opus 4.6 thinking, GPT-OSS 120B - Cursor IDE (
cu): an imported Cursor token; models include Claude 4.6 Opus Max and GPT-5.3 Codex - xAI (Grok) (
xai): OAuth or an API key - Grok CLI (
gcli): uses Grok Build subscription credits, not xAI API credits - Kilo Code, Cline, ClinePass and Kimchi
Claude Code, Codex, GitHub Copilot and Antigravity, and Kiro, Gemini CLI and Qoder below, use a subscription or OAuth session that is not licensed for proxy or router use. The dashboard shows a risk notice and asks you to confirm before connecting: the account may be restricted or banned. Read each provider's terms before you rely on it.
Free
The dashboard lists six providers as free. Four need a sign-in; OpenCode Free and MiMo Code Free need nothing at all:
- Kiro AI (
kr): device-code login; Claude Sonnet 4.5 and Haiku 4.5, GLM-5, MiniMax M2.5, DeepSeek 3.2 and more - Gemini CLI (
gc): Google says that for unpaid-tier and Google One users, Gemini CLI was replaced by Antigravity CLI on June 18, 2026. kRouter still lists the provider; see what changed for Gemini CLI's free tier - Qoder (
qd) - CodeBuddy CN (
cbcn): OAuth - OpenCode Free (
oc): no sign-in; the model list comes from OpenCode - MiMo Code Free (
mmf): no sign-in; servesmimo-auto
Whether a service still has a free quota, and how big it is, is decided by that service and changes often. Check its own page.
Free tier
An API key from a platform with a free allowance or starter credits: Atomesus, Poolside, OpenRouter, NVIDIA NIM, Ollama Cloud, Vertex AI, Gemini (an AI Studio key), Cloudflare Workers AI and BytePlus ModelArk. Each platform sets its own allowance; the provider's page in the dashboard links to where you get the key.
API key (pay per token)
Paste a key and pay the provider directly. This covers OpenAI, Anthropic, DeepSeek, GLM (Z.ai), Kimi, MiniMax, OpenCode Go, Groq, Mistral, Perplexity, Perplexity Agent, Together AI, Fireworks, Cerebras, Cohere, SiliconFlow, Nebius, Chutes, Hyperbolic, Featherless, Venice AI, Azure OpenAI, Vercel AI Gateway and more, plus speech, image, embedding and search services (see Multimodal & Agent Skills).
OpenCode Go serves its models on three protocols. kRouter picks the right one per model, and you can override it per model; see the Responses API section and OpenCode Go's three protocols.
Browser cookie
Paste a session cookie from your browser: Grok Web (gw) and Perplexity Web (pw, Pro or Max).
Your own endpoints
Add any OpenAI-compatible or Anthropic-compatible base URL, or a custom embedding endpoint, as a provider of its own. A model server on your own machine or network works this way, and there is a dedicated Ollama Local provider.
Connecting a provider
- Open the dashboard at
http://localhost:20128/dashboard - Go to Providers and choose the provider
- Add a connection. OAuth providers open a sign-in page or show a device code; API-key providers ask for the key; OpenCode Free and MiMo Code Free need nothing
- Press Test on one of its models to check that it answers
Provider aliases
Model names are always <alias>/<model-id>. Everything before the first / names the provider (its alias or its full ID), and the rest is passed on as the model ID, slashes included. A name without a / is treated as a combo or a model alias.
| Provider | Alias | Example model |
|---|---|---|
| Claude Code | cc | cc/claude-opus-4-7 |
| OpenAI Codex | cx | cx/gpt-5.5 |
| GitHub Copilot | gh | gh/claude-sonnet-4.6 |
| Antigravity | ag | ag/gemini-pro-agent |
| Kiro AI | kr | kr/claude-sonnet-4.5 |
| OpenCode Go | ocg | ocg/kimi-k2.6 |
| GLM Coding (Z.ai) | glm | glm/glm-5.1 |
| DeepSeek | ds | ds/deepseek-v4-pro |
| OpenRouter | openrouter | openrouter/ followed by OpenRouter's own model ID |
The Providers page lists every provider with example models.
Several accounts per provider
You can connect more than one account of the same provider. kRouter picks one for each request: by default with Zenith scoring, or with round-robin, P2C or random. The global choice is under Settings, and each provider's page can override it. Accounts that are rate limited or blocked sit out until they recover. See The Zenith Routing Engine.
Spreading requests across your own accounts does not change a provider's terms. If a provider forbids sharing or rotating accounts, using kRouter does not make it allowed.
Where to go next
- Getting Started: install and connect a first provider
- Combos & fallback: stack providers into fallback chains
- API Reference: the endpoints kRouter serves
- Browse all providers: the searchable list