ask-agent 0.8.1 → 0.9.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9a50ba3d8610e0e05d54ff53c23757a4ef6ecfddb3bc43094d469ebd252e6cfa
-  data.tar.gz: 2269528d25df7991e2aefbc84c4fc89a5a6a4ac4b1d9fbfb95ab9dd421bfa9ba
+  metadata.gz: edea591926e0360d2eb24990d423ef183e3abf99bc141df768a7ee11774db288
+  data.tar.gz: 2e5f013ee517281679e0ab717792ce3110411389b1b2df185b7d6c4c7636e106
 SHA512:
-  metadata.gz: cc258ca1b71919c61a5986f18190c3c5b2bed2d17d1f8efa6e63a8f8f37e258694603ce9b9c382e1843b6c03f94f7121b976f2875d9e429146bd024dfbd9162c
-  data.tar.gz: b371ea561dcd8d67e49c26e0fa18eaa0d142040f0e1adb4de7eba9c517d5cc7e03892dda1edccf2f0b46a3c2e672408f4fe8103547dee9ebd029f5d6f23d0482
+  metadata.gz: d7a430b5c791d0d25920836d604f529b1cac16e8c608f6ebaf1c04b0e5b449a1887c8b3a293ed1e2c6e75cdea23fc7c8614f0e80860c6cfad234951eaf8bc1f7
+  data.tar.gz: 580ccbb6e82210c446f7c8669d0b57a3fb90297156502d6a6c4a872c3f871c6838722c754fe02577083dc49f5cd28da9365d5f9da800d5e1921302bd8bd10958

data/CHANGELOG.md

@@ -1,3 +1,31 @@
+## [0.9.1] — 2026-07-21
+
+### Changed
+
+- **`prompt_caching` now defaults to `true`** globally. All sessions automatically send cache-control hints to supporting providers (Anthropic, OpenAI). Non-supporting providers ignore the parameter safely.
+
+## [0.9.0] — 2026-07-21
+
+### Added
+
+- **Prompt caching support** — `prompt_caching` option enables provider-native prompt caching for significant cost savings on repeated conversation prefixes. Works with both Anthropic and OpenAI.
+
+  ```ruby
+  # Global config
+  Ask::Agent.configure do |c|
+    c.prompt_caching = true
+  end
+
+  # Or per-session
+  session = Ask::Agent::Session.new(model: "claude-sonnet-4", prompt_caching: true)
+  ```
+
+  **Anthropic**: Caches the system prompt and the last user message content. The provider automatically returns cached reads instead of processing the full context on repeated calls. Response metadata includes `cache_creation_input_tokens` and `cache_read_input_tokens`.
+
+  **OpenAI**: Caching is automatic for prompts exceeding 1024 tokens. Response metadata includes `cached_tokens` from `usage.prompt_tokens_details.cached_tokens`.
+
+- **Prompt caching capability** — Both `Ask::Providers::Anthropic` and `Ask::Providers::OpenAI` now advertise `prompt_caching: true` in their capabilities.
+
 ## [0.8.1] — 2026-07-21
 
 ### Added

data/lib/ask/agent/chat.rb

@@ -33,7 +33,7 @@ module Ask
 
       attr_writer :test_provider
 
-      def initialize(model:, tools: [], temperature: nil, schema: nil, provider: nil, **)
+      def initialize(model:, tools: [], temperature: nil, schema: nil, provider: nil, prompt_caching: nil, **)
         @model_id = model.respond_to?(:id) ? model.id : model.to_s
         @model_info = Ask::ModelCatalog.find(@model_id)
         @tools = tools
@@ -43,10 +43,11 @@ module Ask
         @provider_override = provider
         @provider = nil
 
-        # Read configured middleware and stream transforms from global config
+        # Read configured middleware, transforms, and caching from global config
         config = Ask::Agent.configuration
         @middleware_pipeline = config.middleware.configured? ? config.middleware : nil
         @transform_pipeline = config.stream_transforms.configured? ? config.stream_transforms : nil
+        @prompt_caching = prompt_caching.nil? ? config.prompt_caching : prompt_caching
       end
 
       def ask(message = nil, &block)
@@ -142,6 +143,9 @@ module Ask
       # Build the request hash for the provider call.
       # Middleware can mutate this hash to inject defaults, modify params, etc.
       def build_request(stream)
+        extra = (@extra_params || {}).dup
+        extra[:prompt_caching] = true if @prompt_caching
+
         {
           messages: @messages.map(&:to_h),
           model: @model_id,
@@ -149,7 +153,7 @@ module Ask
           temperature: @temperature,
           stream: stream,
           schema: @schema&.respond_to?(:to_json_schema) ? @schema.to_json_schema : @schema,
-          extra_params: (@extra_params || {}).dup
+          extra_params: extra
         }
       end
 

data/lib/ask/agent/configuration.rb

@@ -19,7 +19,8 @@ module Ask
         @compactor_enabled = true
         @compactor_threshold = 0.8
         @parallel_tool_execution = true
-        @prompt_caching = false
+        @max_tool_retries = 3
+        @prompt_caching = true
 
         @middleware = Middleware::Pipeline.new
         @stream_transforms = StreamTransforms::Pipeline.new

data/lib/ask/agent/version.rb

@@ -2,6 +2,6 @@
 
 module Ask
   module Agent
-    VERSION = "0.8.1"
+    VERSION = "0.9.1"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ask-agent
 version: !ruby/object:Gem::Version
-  version: 0.8.1
+  version: 0.9.1
 platform: ruby
 authors:
 - Kaka Ruto