omniai-anthropic 3.11.0 → 3.12.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c25e922a88c2e1af6329ab3dce2308bcc75d8ce8bb02dfba1e55149aaf427488
-  data.tar.gz: 1eb6c6dcdd6a01265d6e65de7347c05c107f5b9ea0adc782a8dae66a2c9a9d7d
+  metadata.gz: 07dc160091024eb4028a842392bbdf5921c74580217306b0610e0f347fda80b8
+  data.tar.gz: 0afd8a083a99409182777f74baf8ef128f60fc94b27650dda2b5dc10f3334c08
 SHA512:
-  metadata.gz: b3cf984aa4edb6f096eb2c9e61e3f94f11ea0ab602c7aec06da671ea39747778c9e653a08f1bf006ace435bf8439f5a81379b82e24771ce6873b088f7ef9e68d
-  data.tar.gz: ef55f2e7b1bdc99983a7d3b04ead43b0528d2788ffb79d18b61ce84b8736036f272728ce220e12d2611c1302d7a214dbf8c6047fa7f2dd739d2acaec8ac87593
+  metadata.gz: e18b6a471f4113af70d6c6dad210fd150d3502f68f812f7f30fd16f4cd30071dcf52bf2ae4dbfb0aec4d729e12dbc30e2fd76c4a13c9853bfe1ab6ac6b0dcb61
+  data.tar.gz: e41208c9d2904ec7a2f94af28ecb28547a827e6d34fbb5a6f0f31b5a56be284a28d893306cd8750d75f34ea0ac65012b707333701abc8cae9eb48fa48f3f09c5

data/README.md

@@ -180,6 +180,12 @@ client.chat(prompt, tools:, cache: true) # 5-minute TTL
 client.chat(prompt, tools:, cache: { ttl: "1h" }) # 1-hour TTL
 ```
 
+For single-shot prompts that share a system prompt but send a different message each time (e.g. one call per document), `cache: :system` marks only the system prompt (or the last tool), so the unique message is not written to the cache:
+
+```ruby
+client.chat(prompt, cache: :system) # 5-minute TTL
+```
+
 `usage.input_tokens` reports the whole prompt, including cached tokens. The cache breakdown is on each response's raw usage:
 
 ```ruby

data/lib/omniai/anthropic/chat.rb

@@ -186,12 +186,12 @@ module OmniAI
       end
 
       # When caching, the last block of the last message is marked so each tool-loop round reads the history the
-      # previous round wrote.
+      # previous round wrote. `cache: :system` skips this for single-shot prompts whose message is never reused.
       #
       # @return [Array<Hash>]
       def messages
         messages = @prompt.messages.reject(&:system?).map { |message| message.serialize(context:) }
-        return messages unless cache_control && messages.any?
+        return messages unless cache_control && @options[:cache] != :system && messages.any?
 
         *history, last = messages
         history + [last.merge(content: with_cache_control(last[:content]))]
@@ -211,12 +211,13 @@ module OmniAI
 
       # Translates the opt-in `cache` option to Anthropic's `cache_control`.
       # Example: `cache: true` becomes `{ type: "ephemeral" }` (5-minute TTL)
+      # Example: `cache: :system` becomes `{ type: "ephemeral" }` on the system prompt (or last tool) only
       # Example: `cache: { ttl: "1h" }` becomes `{ type: "ephemeral", ttl: "1h" }`
       #
       # @return [Hash, nil]
       def cache_control
         case @options[:cache]
-        when true then { type: "ephemeral" }
+        when true, :system then { type: "ephemeral" }
         when Hash then { type: "ephemeral" }.merge(@options[:cache])
         end
       end

data/lib/omniai/anthropic/version.rb

@@ -2,6 +2,6 @@
 
 module OmniAI
   module Anthropic
-    VERSION = "3.11.0"
+    VERSION = "3.12.0"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: omniai-anthropic
 version: !ruby/object:Gem::Version
-  version: 3.11.0
+  version: 3.12.0
 platform: ruby
 authors:
 - Kevin Sylvestre