How do you run Kimi K3 in Claude Code?

Claude Code sends every request in Anthropic's Messages format. It reaches another server when ANTHROPIC_BASE_URL is set to that server's address.

Moonshot AI serves Kimi K3 in that same format, at https://api.moonshot.ai/anthropic. OpenRouter and Ollama Cloud answer in it too.

So unlike GPT, Kimi K3 needs no translation in the middle. What it needs is the right model names, the right window and a key. How GPT differs is in running GPT in Claude Code.

Moonshot's own guide puts ten variables in the env block of ~/.claude/settings.json:

• the base URL above, and your Moonshot API key as ANTHROPIC_AUTH_TOKEN
• kimi-k3[1m] as the main model, and as the Opus, Sonnet, Fable and subagent model
• kimi-k2.7-code as the Haiku model
• CLAUDE_CODE_AUTO_COMPACT_WINDOW at 1000000, and the effort level at max

Moonshot warns that subagent work fails or degrades noticeably without the subagent variable. It also warns that the file now holds your API key in plain text, so keep it out of git.

That works for one person. On a team, it is one Moonshot key and one settings file per laptop, and no shared view of what Kimi costs. Why a key on every laptop is the expensive part is in keeping API keys off laptops.

An owner connects an OpenRouter key or an Ollama Cloud plan in Selan. Kimi K3 is a model Selan lists, so an owner also switches it on at OpenRouter under Settings → Models.

Developers install the selan CLI, sign in as themselves and name the model:

The part before the first colon is the provider. Everything after it is that provider's own model id, passed through as it is.

With --model, Selan tells Claude Code the model's real context window. Selan lists Kimi K3's window as 1,048,576 tokens, so a session compacts there and not at 200K. If an owner set a lower context limit for your company, it compacts at that instead.

Kimi K3 starts at max effort, the level Moonshot recommends. Selan offers low, high and max for it.

To skip the flag, an owner adds Kimi K3 to Claude Code's picker under Settings → Model pickers. It then appears in /model. The details are in Run any model.

If that account is out of quota, the request fails rather than moving to the other one.

Ollama answers only when you name one of its models. A request that names no model never lands on it. All of this is on Providers and models.

• One usage record. Kimi turns land beside Claude turns, per person and per repository. Usage per developer shows what an owner sees.
• One spend limit. An OpenRouter turn is metered, so it counts against a member's limit. A turn on an Ollama flat plan does not. See spend limits.
• No Moonshot or OpenRouter key on a laptop. Keys stay on the gateway. A developer holds a Selan session you can revoke.
• Codex on the same accounts. selan codex --model openrouter:moonshotai/kimi-k3 uses the same route. See Codex.

• Effort and spend. Kimi K3 thinks before it answers, and Selan starts it at max. On OpenRouter, the thinking is metered too.
• The model must be switched on. A model your company cannot use is refused before Claude Code starts, not at the first request.

Free for up to three seats, and the Free plan needs no card.