whoosh 1.9.0 → 1.10.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b74890ede8ed95aadeafdd55439b0a267ec521ecb5c84fd879807f29789a9d9a
-  data.tar.gz: aa04e3dbf492846d8f06fe057ef657a137065a6b8069421c24013a4102b63511
+  metadata.gz: 0c8ecbe08e5ecf50f7f1233da3767074a79d9e4164a719e9397be1ac9637efb3
+  data.tar.gz: 622f3f889c550ecdff4cfab4f38120f2ce3f14b7f2021be99dae9790c3cfc5f7
 SHA512:
-  metadata.gz: 884921ba3182cef85545192b08301ced1d992705621ce0860e55e916da2ece5f8d1b122f676a88e0e7b18dcbb9f271e01e2e46bbd5439962cd476bfb73173078
-  data.tar.gz: 8dcb16978fa00228a18de47e42cbe29c0e3d8da84c3bf3a6b4b52842ea9ce4cbcc7a5930b1c471c8b2d06065b6793db6574a8ec2392e7f9707e07d73940be0fb
+  metadata.gz: 70add9595209f2ff3a1c5c4a52e25a4428149694f4df9debc7cc58dace2d15b77d46dcf1591cb9cd78e7d02654f0402326c58c9e060f1c94897e0011fc2cfab2
+  data.tar.gz: 83fb93d35d3b1d446bc99d1f8ee0a21b16f062af81da4f4e737b98cd85acced622fe332ccefb785c5f114fa78131131bdea9f008e109c49291c793c117cea843

data/lib/whoosh/ai/llm.rb

@@ -2,15 +2,48 @@
 
 module Whoosh
   module AI
+    # Bounded LRU cache. Ruby's Hash preserves insertion order, so we reorder
+    # on read (delete+reinsert) and evict the oldest entry when over capacity.
+    class LRUCache
+      def initialize(max_size)
+        @max_size = max_size
+        @store = {}
+        @mutex = Mutex.new
+      end
+
+      def [](key)
+        @mutex.synchronize do
+          return nil unless @store.key?(key)
+          value = @store.delete(key)
+          @store[key] = value
+        end
+      end
+
+      def []=(key, value)
+        @mutex.synchronize do
+          @store.delete(key) if @store.key?(key)
+          @store[key] = value
+          @store.shift while @store.size > @max_size
+          value
+        end
+      end
+
+      def size
+        @store.size
+      end
+    end
+
+    DEFAULT_MODEL     = "claude-sonnet-4-6"
+    DEFAULT_CACHE_MAX = 1000
+
     class LLM
       attr_reader :provider, :model
 
-      def initialize(provider: "auto", model: nil, cache_enabled: true)
+      def initialize(provider: "auto", model: nil, cache_enabled: true, cache_size: DEFAULT_CACHE_MAX)
         @provider = provider
         @model = model
         @cache_enabled = cache_enabled
-        @cache = cache_enabled ? {} : nil
-        @mutex = Mutex.new
+        @cache = cache_enabled ? LRUCache.new(cache_size) : nil
         @ruby_llm = nil
       end
 
@@ -32,10 +65,7 @@ module Whoosh
           temperature: temperature
         )
 
-        # Cache result
-        if use_cache && @cache
-          @mutex.synchronize { @cache[cache_key] = result }
-        end
+        @cache[cache_key] = result if use_cache && @cache
 
         result
       end
@@ -60,18 +90,19 @@ module Whoosh
         end
       end
 
-      # Stream LLM response — yields chunks
+      # Stream an LLM response — yields each chunk as ruby_llm produces it.
+      # The block receives a RubyLLM::Chunk (a Message subclass with #content).
+      # Returns the final response message after the stream completes.
       def stream(message, model: nil, system: nil, &block)
         ensure_ruby_llm!
+        unless @ruby_llm
+          raise Errors::DependencyError, "No LLM provider available. Add 'ruby_llm' to your Gemfile."
+        end
 
-        messages = [{ role: "user", content: message }]
-        # Delegate to ruby_llm's streaming interface
-        if @ruby_llm
-          # ruby_llm streaming would go here
-          # For now, fall back to non-streaming
-          result = chat(message, model: model, system: system, cache: false)
-          yield result if block_given?
-          result
+        chat = RubyLLM.chat(model: model || @model || DEFAULT_MODEL)
+        chat.with_instructions(system) if system
+        chat.ask(message) do |chunk|
+          block.call(chunk) if block
         end
       end
 
@@ -87,7 +118,7 @@ module Whoosh
 
         if @ruby_llm
           # Use ruby_llm gem
-          chat = RubyLLM.chat(model: model || "claude-sonnet-4-20250514")
+          chat = RubyLLM.chat(model: model || DEFAULT_MODEL)
           chat.with_instructions(system) if system
           response = chat.ask(messages.last[:content])
           response.content

data/lib/whoosh/ai.rb

@@ -11,7 +11,8 @@ module Whoosh
       LLM.new(
         provider: ai_config["provider"] || "auto",
         model: ai_config["model"],
-        cache_enabled: ai_config["cache"] != false
+        cache_enabled: ai_config["cache"] != false,
+        cache_size: ai_config["cache_size"] || DEFAULT_CACHE_MAX
       )
     end
   end

data/lib/whoosh/streaming/llm_stream.rb

@@ -17,7 +17,8 @@ module Whoosh
 
       def <<(chunk)
         return if @closed
-        text = chunk.respond_to?(:text) ? chunk.text : chunk.to_s
+        text = extract_text(chunk)
+        return self if text.nil? || text.empty?
         payload = { choices: [{ delta: { content: text } }] }
         write("data: #{JSON.generate(payload)}\n\n")
         self
@@ -40,6 +41,21 @@ module Whoosh
 
       private
 
+      # ruby_llm chunks expose #content (Message subclass). Older code paths
+      # and plain-string yields are also supported.
+      def extract_text(chunk)
+        return chunk if chunk.is_a?(String)
+        if chunk.respond_to?(:content)
+          c = chunk.content
+          return "" if c.nil?
+          return c if c.is_a?(String)
+          return c.text if c.respond_to?(:text)
+          return c.to_s
+        end
+        return chunk.text if chunk.respond_to?(:text)
+        chunk.to_s
+      end
+
       def write(data)
         @io.write(data)
         @io.flush if @io.respond_to?(:flush)

data/lib/whoosh/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Whoosh
-  VERSION = "1.9.0"
+  VERSION = "1.10.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: whoosh
 version: !ruby/object:Gem::Version
-  version: 1.9.0
+  version: 1.10.0
 platform: ruby
 authors:
 - Johannes Dwi Cahyo