How do you use Ollama with 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.

Ollama answers in that format from version 0.14.0. Its local server does, and so does its cloud at https://ollama.com.

Ollama supports a subset of the format. Messages, streaming and tool calls work on compatible models. Tool-choice controls, deferred tools and hosted web search are not fully supported.

Install Ollama and Claude Code, then start Claude Code through Ollama:

To set it up yourself, Ollama's docs give two variables. ANTHROPIC_BASE_URL is http://localhost:11434. ANTHROPIC_AUTH_TOKEN is ollama, which is required but ignored.

Pick a model trained for tool calls, because Claude Code edits files and runs commands through tools. For a larger repository, Ollama says to set the context length to 64k or higher.

The model then runs on your own hardware, and Ollama's plans put no limit on local use. How fast it answers depends on that hardware.

Ollama's cloud needs no Ollama install. Create an API key on ollama.com, then start Claude Code against it:

The cloud takes the key only as a bearer token, so it goes in ANTHROPIC_AUTH_TOKEN. Claude Code's settings files override these shell variables, so change them there too, and /status checks the connection.

Name a cloud model as https://ollama.com/api/tags lists it, such as gemma4:31b. The Ollama app and CLI spell the same model gemma4:cloud.

Ollama's paid plans include a monthly amount of usage. It is measured in tokens at each model's rate, and use past it draws on credits you buy. Ollama is one account per person, and its Team plan is the one sold for shared use.

That works for one person. On a team, it is one Ollama key and one set of variables per laptop, and no shared view of who used what. Why a key on every laptop is the expensive part is in keeping API keys off laptops.

An owner pastes the company's Ollama API key in Selan, then switches on the Ollama models the team may use under Settings → Models. Selan lists Gemma 4, DeepSeek V4.1 Flash and DeepSeek V4 Pro on Ollama by name.

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 Ollama's own model id, passed through as it is.

With --model, Selan tells Claude Code the model's real context window. Selan lists Gemma 4 at 262,144 tokens, DeepSeek V4.1 Flash at 1,048,576 and V4 Pro at 1,000,000. If an owner set a lower context limit for your company, a session compacts there instead.

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

• No local models. To run a model on your own machine, use Ollama's own setup above. Selan's usage record and spend limits do not see those turns.
• No silent swaps. Ollama answers only when you name one of its models. A request that names no model never lands on it.
• No fallback. An Ollama turn runs on Ollama and on no other provider. If the key is out of usage, the request fails rather than moving elsewhere.
• No spend-limit cap. Selan has no per-token price for Ollama, so an Ollama turn does not count against a member's spend limit. Ollama's own usage page and credit settings cap that use.

Every provider Selan connects is on Providers and models.

• One usage record. Ollama turns land beside Claude turns, per person and per repository. Usage per developer shows what an owner sees.
• No Ollama key on a laptop. The key stays on the gateway. A developer holds a Selan session you can revoke.
• Codex on the same key. selan codex --model ollama:gemma4:31b uses Ollama's own Responses API. Codex edits files through the shell there. See Codex.
• Other models beside it. The same session can name Claude, GPT or Kimi on your other accounts. See any model in Claude Code.

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