Elixir client for the Claude API.

Build a client with new/1, then hand it to the module for whatever you're doing:

client = Claudex.new(api_key: "sk-ant-...")

{:ok, message} =
  Claudex.Messages.create(client, %{
    model: "claude-opus-5",
    max_tokens: 1024,
    messages: [%{role: "user", content: "Hello, Claude"}]
  })

Claudex.Message.text(message)

A client is a plain struct, so you can hold several at once and pass them around freely. Nothing here runs as a process or reads global state.

What's here

• Claudex.Client - connection settings: API key, base URL, timeouts, retries.
• Claudex.Messages - the Messages API. create/2 for a complete reply, stream!/2 and stream_to/3 to get it as it's written, and count_tokens/2 to price a request before sending it.
• Claudex.Stream - assemble a stream back into a message, or cancel one. Claudex.Stream.Event has the structs you match on while it runs.
• Claudex.Tool - turn a module's functions into tools Claude can call, with the JSON schema derived from their @spec and @doc.
• Claudex.ToolRunner - run a whole tool conversation: send, run what Claude asks for, send the results back, repeat.
• Claudex.Models - which models your key can use, and what each supports.
• Claudex.Files - upload a file once, then reference it by file_id instead of re-sending the bytes; also how you download what Claude creates.
• Claudex.Messages.Batches - send up to 100,000 requests at once for asynchronous processing, at half the token cost.
• Claudex.Message, Claudex.ContentBlock, Claudex.Usage - what a reply is made of.
• Claudex.Error - every failure, with a :type you can match on.
• Claudex.Telemetry - the events the SDK emits, and a one-line logger for them when you want to see what it's doing.

Fallible functions return {:ok, result} or {:error, %Claudex.Error{}}. The exceptions are the ! variants, which raise instead.