ask-agent 0.8.1 → 0.9.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9a50ba3d8610e0e05d54ff53c23757a4ef6ecfddb3bc43094d469ebd252e6cfa
-  data.tar.gz: 2269528d25df7991e2aefbc84c4fc89a5a6a4ac4b1d9fbfb95ab9dd421bfa9ba
+  metadata.gz: 82d7ec2cb06f4ce9cbaa41ece0ac4800772b44dcbe54c33e46f6edefb863f1b2
+  data.tar.gz: b8b5cfec092ae7b0117f0a4ec3b441becc0ea21c806107cb0ae9c51c5b215618
 SHA512:
-  metadata.gz: cc258ca1b71919c61a5986f18190c3c5b2bed2d17d1f8efa6e63a8f8f37e258694603ce9b9c382e1843b6c03f94f7121b976f2875d9e429146bd024dfbd9162c
-  data.tar.gz: b371ea561dcd8d67e49c26e0fa18eaa0d142040f0e1adb4de7eba9c517d5cc7e03892dda1edccf2f0b46a3c2e672408f4fe8103547dee9ebd029f5d6f23d0482
+  metadata.gz: d9a3dd55f214105d58858f31122a0064a9a94165f37d5b8fc8d1f202ce87fe19ff8ad4f0c6e52de1d66bf115940e3c48e8673762d16912514668b82cabd8daa3
+  data.tar.gz: ecaecabad77df46494a9ee0f99840cbeeaf59163cecaf1ccb8af8810e98bfa8333455c1ed3b50d1bc66c106c38e74b98329d28fea3d2e4393cacdd9453213c23

data/CHANGELOG.md

@@ -1,3 +1,25 @@
+## [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

@@ -19,6 +19,7 @@ module Ask
         @compactor_enabled = true
         @compactor_threshold = 0.8
         @parallel_tool_execution = true
+        @max_tool_retries = 3
         @prompt_caching = false
 
         @middleware = Middleware::Pipeline.new

data/lib/ask/agent/version.rb

@@ -2,6 +2,6 @@
 
 module Ask
   module Agent
-    VERSION = "0.8.1"
+    VERSION = "0.9.0"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ask-agent
 version: !ruby/object:Gem::Version
-  version: 0.8.1
+  version: 0.9.0
 platform: ruby
 authors:
 - Kaka Ruto