llmemory 0.1.11 → 0.1.13

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 2016468aa862476d2bfc4fafef85e340433b0bf940519351a9fb72f3bda5e796
-  data.tar.gz: 31d1d72a3ef8b94b6cef5e696ed29f72357377670d484897350cddcdf17a20ae
+  metadata.gz: a188251e04aac90f929fcc952f7902a8e9b86728742ab88e5f209953821efd30
+  data.tar.gz: 01f3cd9d68c50a1a52ed33683b973a379073890db1ae5a3073676fd598581eca
 SHA512:
-  metadata.gz: f360315f571e5641f53af4006f661a9e432f80280d53a2a7f99d7563f318f23b04950925d2bb2e0716bcad6088b5f929fa15d2b25523f7d83af04b0ac269db3b
-  data.tar.gz: aef3ce481555a0f84630207ff0333fe749d841656dd41c1c6449b4e3797cff73ecd033ce1817ff50d8e87913e783dfed3198eb85ccadfd6683388f9a4e5a8780
+  metadata.gz: e24d411ffaca985dc360bc6a6d271f8fd57c687cef26c8242e047fab6bcda400b3764b78aeea8d58ec253bc33ce48b89a46d6a8bdb71db8152c4d175880f3b87
+  data.tar.gz: d7097e4c6cc8442088f5adb0614015b8d9d0541db8f4843e729c8e94cfefe0faf49969e9abb75381b070386c53623631b937fc72989a1f7cb0aa27b3b45a9171

data/README.md

@@ -33,6 +33,9 @@ context = memory.retrieve("¿Qué preferencias tiene el usuario?", max_tokens: 2
 # Optionally consolidate current conversation into long-term (extract facts)
 memory.consolidate!
 
+# Compact short-term memory when it gets too large (summarizes old messages)
+memory.compact!(max_bytes: 8192)  # or use config default
+
 # Clear session (short-term) while keeping long-term intact
 memory.clear_session!
 ```
@@ -41,6 +44,7 @@ memory.clear_session!
 - **`messages`** — Returns the current conversation history.
 - **`retrieve(query, max_tokens: nil)`** — Returns combined context: recent conversation + relevant long-term memories.
 - **`consolidate!`** — Extracts facts from the current conversation and stores them in long-term.
+- **`compact!(max_bytes: nil)`** — Compacts short-term memory by summarizing old messages when byte size exceeds limit. Uses LLM to create a summary, keeping recent messages intact.
 - **`clear_session!`** — Clears short-term only.
 
 ## Configuration
@@ -59,6 +63,7 @@ Llmemory.configure do |config|
   config.time_decay_half_life_days = 30
   config.max_retrieval_tokens = 2000
   config.prune_after_days = 90
+  config.compact_max_bytes = 8192  # max bytes before compact! triggers
 end
 ```
 
@@ -263,6 +268,20 @@ The dashboard uses your existing `Llmemory.configuration` (short-term store, lon
 
 llmemory includes an MCP server that allows LLM agents (like Claude Code) to interact directly with the memory system. This gives agents "agency" over their own memory—they can search, save, and retrieve memories autonomously.
 
+### Installation
+
+The MCP server requires the `mcp` gem, which is **optional**. Install it separately:
+
+```bash
+gem install mcp
+```
+
+Or add to your Gemfile:
+
+```ruby
+gem "mcp", "~> 0.6"
+```
+
 ### Starting the Server
 
 **Stdio mode** (default, for local use with Claude Code):

data/lib/llmemory/cli/commands/mcp.rb

@@ -88,6 +88,17 @@ module Llmemory
         def run_server
           require_relative "../../mcp"
 
+          unless Llmemory::MCP.available?
+            $stderr.puts "Error: The 'mcp' gem is required for MCP server functionality."
+            $stderr.puts ""
+            $stderr.puts "Install it with:"
+            $stderr.puts "  gem install mcp"
+            $stderr.puts ""
+            $stderr.puts "Or add to your Gemfile:"
+            $stderr.puts "  gem 'mcp', '~> 0.6'"
+            exit 1
+          end
+
           server = Llmemory::MCP::Server.new(name: @server_name)
 
           if @http_mode

data/lib/llmemory/configuration.rb

@@ -15,7 +15,8 @@ module Llmemory
                   :vector_store,
                   :time_decay_half_life_days,
                   :max_retrieval_tokens,
-                  :prune_after_days
+                  :prune_after_days,
+                  :compact_max_bytes
 
     def initialize
       @llm_provider = :openai
@@ -32,6 +33,7 @@ module Llmemory
       @time_decay_half_life_days = 30
       @max_retrieval_tokens = 2000
       @prune_after_days = 90
+      @compact_max_bytes = 8192
     end
   end
 

data/lib/llmemory/mcp.rb

@@ -1,8 +1,29 @@
 # frozen_string_literal: true
 
-require_relative "mcp/server"
-
 module Llmemory
   module MCP
+    class << self
+      def available?
+        @available
+      end
+
+      def require_mcp!
+        unless available?
+          raise LoadError, <<~MSG
+            The 'mcp' gem is required for MCP server functionality.
+            Install it with: gem install mcp
+            Or add to your Gemfile: gem 'mcp', '~> 0.6'
+          MSG
+        end
+      end
+    end
+
+    begin
+      require "mcp"
+      @available = true
+      require_relative "mcp/server"
+    rescue LoadError
+      @available = false
+    end
   end
 end

data/lib/llmemory/memory.rb

@@ -52,12 +52,77 @@ module Llmemory
       true
     end
 
+    def compact!(max_bytes: nil)
+      max = max_bytes || Llmemory.configuration.compact_max_bytes
+      msgs = messages
+      current_bytes = messages_byte_size(msgs)
+      return false if current_bytes <= max
+
+      old_msgs, recent_msgs = split_messages_by_bytes(msgs, max)
+      return false if old_msgs.empty?
+
+      summary = summarize_messages(old_msgs)
+      compacted = [{ role: :system, content: summary }] + recent_msgs
+      save_state(messages: compacted)
+      true
+    end
+
     def user_id
       @user_id
     end
 
     private
 
+    def summarize_messages(msgs)
+      conversation = msgs.map { |m| format_message(m) }.join("\n")
+      prompt = <<~PROMPT
+        Summarize the following conversation into a concise summary that preserves key information, decisions, and context. Write it as a brief narrative (max 200 words).
+
+        Conversation:
+        #{conversation}
+
+        Summary:
+      PROMPT
+      llm_client.invoke(prompt.strip).to_s.strip
+    rescue Llmemory::LLMError
+      msgs.map { |m| format_message(m) }.join("\n")[0..500]
+    end
+
+    def llm_client
+      @llm ||= Llmemory::LLM.client
+    end
+
+    def messages_byte_size(msgs)
+      msgs.sum { |m| message_byte_size(m) }
+    end
+
+    def message_byte_size(msg)
+      role = msg[:role] || msg["role"]
+      content = msg[:content] || msg["content"]
+      role.to_s.bytesize + content.to_s.bytesize
+    end
+
+    def split_messages_by_bytes(msgs, max_bytes)
+      target_recent_bytes = max_bytes / 2
+      recent_bytes = 0
+      split_index = msgs.size
+
+      (msgs.size - 1).downto(0) do |i|
+        msg_bytes = message_byte_size(msgs[i])
+        if recent_bytes + msg_bytes <= target_recent_bytes
+          recent_bytes += msg_bytes
+          split_index = i
+        else
+          break
+        end
+      end
+
+      split_index = [split_index, msgs.size - 1].min
+      split_index = [split_index, 1].max if msgs.size > 1
+
+      [msgs[0...split_index], msgs[split_index..]]
+    end
+
     def build_long_term(long_term_type)
       llm_opts = @llm ? { llm: @llm } : {}
       case long_term_type.to_s.to_sym

data/lib/llmemory/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Llmemory
-  VERSION = "0.1.11"
+  VERSION = "0.1.13"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llmemory
 version: !ruby/object:Gem::Version
-  version: 0.1.11
+  version: 0.1.13
 platform: ruby
 authors:
 - llmemory
@@ -30,7 +30,7 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: '0.6'
-  type: :runtime
+  type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements: