ask-agent 0.3.0 → 0.3.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4812b86203eb5a12eb03138de1e50fe8f0a80532f545bb683dcf051f2729c088
-  data.tar.gz: ba9e999b2aedae3e9c0f55067a35e3d3f29f828717a8f7129c0e0037bd518a33
+  metadata.gz: 26c124febbf2652616754dd539cc8df5d24afe5fbe8c03381f760598578a051f
+  data.tar.gz: 50d87794d60901a85da426f50a1a619807faf96bbe1fedb58145dd2d5d7fbeb1
 SHA512:
-  metadata.gz: a5b049b05acf1a82334a2169a708d60a6511db2d790ec5db3f30b9b4775a561eb204b8fdb209b60d095edeba43bd27f1572a909d78118eba9f295943625da370
-  data.tar.gz: 54ba99bbacd6152c52196f276f28559d2a13e09b048338c4c5198e9ab01c8f2cc9e5eb57c2a957b1ba6dea4410726c3eba644104b66d767729e908a10c366133
+  metadata.gz: 5927098dcbd1fe84faa0cf2a7cddbc63e544a7113e47c61f972044e598053758417edce0e63d9a55d7922fbbc65ad9f552ef5c3fb1c07ac71ba1b718d1cbb815
+  data.tar.gz: 0436ac54f1dcd57ac06706d7e6a8ad5030574341170e0c7e99ca174d3dd94341651142c78bf5d138be1f99a978d71eb21f03a57df1ef4b6cb038ebb22778a637

data/CHANGELOG.md

@@ -1,3 +1,13 @@
+## [0.3.1] — 2026-07-17
+
+### Added
+
+- **Rate-limit aware retry in Chat** — `Chat#ask` retries up to 3 times on `RateLimitError`, using `retry_after` from the error when available, otherwise exponential backoff with jitter.
+
+### Fixed
+
+- **`retryable_error_name?` in ToolExecutor** — fixed duplicate `Ask::RateLimitError` and non-existent `Ask::ServiceUnavailableError`. Now uses class hierarchy matching so subclasses are also retried. (Backport from LiteLLM error classification.)
+
 ## [0.3.0] — 2026-07-17
 
 ### Added

data/lib/ask/agent/chat.rb

@@ -49,27 +49,7 @@ module Ask
         provider_schema = @schema&.respond_to?(:to_json_schema) ? @schema.to_json_schema : @schema
         provider_params = @extra_params || {}
 
-        result = provider.chat(
-          @messages.map(&:to_h),
-          model: provider_model,
-          tools: provider_tools,
-          temperature: provider_temp,
-          stream: stream,
-          schema: provider_schema,
-          **provider_params
-        ) do |raw_chunk|
-          next unless block_given?
-
-          accumulate_tool_calls(raw_chunk, calls_acc)
-
-          yield ChatChunk.new(
-            content: raw_chunk.content,
-            tool_calls: build_current_tool_calls(calls_acc),
-            thinking: raw_chunk.respond_to?(:thinking) ? raw_chunk.thinking : nil,
-            input_tokens: nil,
-            output_tokens: nil
-          )
-        end
+        result = chat_with_retry(stream, calls_acc, &block)
 
         response_msg = if stream
           build_stream_response(result, calls_acc)
@@ -234,6 +214,41 @@ module Ask
         nil
       end
 
+      MAX_CHAT_RETRIES = 3
+
+      def chat_with_retry(stream, calls_acc, &block)
+        MAX_CHAT_RETRIES.times do |attempt|
+          begin
+            return provider.chat(
+              @messages.map(&:to_h),
+              model: @model_id,
+              tools: @tools.map { |t| Ask::ToolDef.from_tool(t) },
+              temperature: @temperature,
+              stream: stream,
+              schema: @schema&.respond_to?(:to_json_schema) ? @schema.to_json_schema : @schema,
+              **(@extra_params || {})
+            ) do |raw_chunk|
+              next unless block
+
+              accumulate_tool_calls(raw_chunk, calls_acc)
+
+              block.call(ChatChunk.new(
+                content: raw_chunk.content,
+                tool_calls: build_current_tool_calls(calls_acc),
+                thinking: raw_chunk.respond_to?(:thinking) ? raw_chunk.thinking : nil,
+                input_tokens: nil,
+                output_tokens: nil
+              ))
+            end
+          rescue Ask::RateLimitError => e
+            raise if attempt >= MAX_CHAT_RETRIES - 1
+
+            delay = e.retry_after || ((2 ** attempt) + rand(0.0..1.0))
+            sleep(delay)
+          end
+        end
+      end
+
       def emit_instrumentation(stream, response_msg)
         return unless defined?(Ask::Instrumentation)
 

data/lib/ask/agent/tool_executor.rb

@@ -182,10 +182,15 @@ module Ask
       end
 
       def retryable_error_name?(error_name)
-        retryable = %w[Timeout::Error Errno::ETIMEDOUT
-                       Ask::RateLimitError Ask::ServerError
-                       Ask::RateLimitError Ask::ServiceUnavailableError]
-        retryable.include?(error_name)
+        return false unless error_name
+
+        klass = Object.const_get(error_name) rescue nil
+        return false unless klass
+
+        klass <= Ask::RateLimitError ||
+        klass <= Ask::ServerError ||
+        klass <= Ask::ServiceUnavailable ||
+        %w[Timeout::Error Errno::ETIMEDOUT].include?(error_name)
       end
 
       def critical_error?(error_class_name)

data/lib/ask/agent/version.rb

@@ -2,6 +2,6 @@
 
 module Ask
   module Agent
-    VERSION = "0.3.0"
+    VERSION = "0.3.1"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ask-agent
 version: !ruby/object:Gem::Version
-  version: 0.3.0
+  version: 0.3.1
 platform: ruby
 authors:
 - Kaka Ruto