Add support for prompt caching

For more details: https://www.anthropic.com/news/prompt-caching

src/anthropic.cr

@@ -3,3 +3,4 @@ require "./version"
 require "./messages"
 require "./message"
 require "./model"
+require "./cache_control"

src/cache_control.cr

@@ -0,0 +1,14 @@
+require "./resource"
+
+struct Anthropic::CacheControl
+  include Resource
+
+  getter type : Type
+
+  def initialize(@type = :ephemeral)
+  end
+
+  enum Type
+    Ephemeral
+  end
+end

src/generated_message.cr

@@ -154,6 +154,8 @@ struct Anthropic::GeneratedMessage
 
     getter input_tokens : Int64
     getter output_tokens : Int64
+    getter cache_creation_input_tokens : Int64 = 0
+    getter cache_read_input_tokens : Int64 = 0
   end
 
   enum StopReason

src/message_content.cr

@@ -1,13 +1,17 @@
-require "./resource"
 require "json"
 require "base64"
 
+require "./resource"
+require "./cache_control"
+
 module Anthropic
   abstract struct MessageContent
     include Resource
 
     abstract def type : String
 
+    getter cache_control : CacheControl?
+
     use_json_discriminator "type", {
       text:        Text,
       image:       Image,
@@ -26,7 +30,7 @@ module Anthropic
     getter type : String = "text"
     getter text : String
 
-    def initialize(@text)
+    def initialize(@text, *, @cache_control = nil)
     end
 
     def to_s(io) : Nil
@@ -40,15 +44,16 @@ module Anthropic
     getter type : String = "image"
     getter source : Source
 
-    def self.base64(media_type : Source::MediaType, data : String)
+    def self.base64(media_type : Source::MediaType, data : String, cache_control : CacheControl? = nil)
       new(
         type: :base64,
         media_type: media_type,
         data: Base64.strict_encode(data),
+        cache_control: cache_control,
       )
     end
 
-    def initialize(*, type : Source::Type, media_type : Source::MediaType, data : String)
+    def initialize(*, type : Source::Type, media_type : Source::MediaType, data : String, @cache_control = nil)
       @source = Source.new(type: type, media_type: media_type, data: data)
     end
 

src/messages.cr

@@ -4,6 +4,22 @@ require "./event_source"
 
 module Anthropic
   struct Messages < API
+    CREATE_HEADERS = HTTP::Headers{
+      "anthropic-beta" => {
+        # Not sure if this one is still necessary because tools seem to work
+        # without it, but keeping it here until I see something official.
+        "tools-2024-04-04",
+
+        # 3.5 Sonnet only supports 4k tokens by default, but you can opt into
+        # up to 8k output tokens.
+        # https://x.com/alexalbert__/status/1812921642143900036
+        "max-tokens-3-5-sonnet-2024-07-15",
+
+        # https://www.anthropic.com/news/prompt-caching
+        "prompt-caching-2024-07-31",
+      }.join(','),
+    }
+
     def create(
       *,
       model : String,
@@ -29,19 +45,10 @@ module Anthropic
         tools: tools.to_json,
       )
 
-      headers = HTTP::Headers.new
-
-      # 3.5 Sonnet only supports 4k tokens by default, but you can opt into
-      # up to 8k output tokens.
-      # https://x.com/alexalbert__/status/1812921642143900036
-      if model.includes?("3-5-sonnet") && max_tokens > 4096
-        headers.add "anthropic-beta", "max-tokens-3-5-sonnet-2024-07-15"
-      end
-
       # TODO: Investigate whether the `beta=tools` is needed since we're using
       # the tools beta header above.
       response = client.post "/v1/messages?beta=tools",
-        headers: headers,
+        headers: CREATE_HEADERS,
         body: request,
         as: GeneratedMessage
       response.message_thread = messages.dup << response.to_message