Add sqlite-vec support

Gemfile.lock

@@ -402,6 +402,13 @@ GEM
     spreadsheet (1.3.1)
       bigdecimal
       ruby-ole
+    sqlite3 (1.7.3)
+      mini_portile2 (~> 2.8.0)
+    sqlite3 (1.7.3-aarch64-linux)
+    sqlite3 (1.7.3-arm-linux)
+    sqlite3 (1.7.3-arm64-darwin)
+    sqlite3 (1.7.3-x86_64-darwin)
+    sqlite3 (1.7.3-x86_64-linux)
     standard (1.39.1)
       language_server-protocol (~> 3.17.0.2)
       lint_roller (~> 1.0)
@@ -493,6 +500,7 @@ DEPENDENCIES
   ruby-openai (~> 7.1.0)
   safe_ruby (~> 1.0.4)
   sequel (~> 5.87.0)
+  sqlite3 (~> 1.7.0)
   standard (>= 1.35.1)
   vcr
   weaviate-ruby (~> 0.9.2)

examples/sqlite_vec_example.rb

@@ -0,0 +1,46 @@
+require "langchain"
+
+# Initialize the LLM (using Ollama in this example)
+llm = Langchain::LLM::Ollama.new
+
+# Initialize the SQLite-vec vectorstore
+db = Langchain::Vectorsearch::SqliteVec.new(
+  url: ":memory:", # Use a file-based DB by passing a path or ":memory:" for in-memory
+  index_name: "documents",
+  namespace: "test",
+  llm: llm
+)
+
+# Create the schema
+db.create_default_schema
+
+# Add some sample texts
+texts = [
+  "Ruby is a dynamic, open source programming language with a focus on simplicity and productivity.",
+  "Python is a programming language that lets you work quickly and integrate systems more effectively.",
+  "JavaScript is a lightweight, interpreted programming language with first-class functions.",
+  "Rust is a multi-paradigm, general-purpose programming language designed for performance and safety."
+]
+
+puts "Adding texts..."
+ids = db.add_texts(texts: texts)
+puts "Added #{ids.size} texts with IDs: #{ids.join(", ")}"
+
+# Search for similar texts
+query = "What programming language is focused on memory safety?"
+puts "\nSearching for: #{query}"
+results = db.similarity_search(query: query)
+
+puts "\nResults:"
+results.each do |result|
+  puts "- #{result[1]}"
+end
+
+# Ask a question
+question = "Which programming language emphasizes simplicity?"
+puts "\nAsking: #{question}"
+response = db.ask(question: question)
+puts "Answer: #{response.chat_completion}"
+
+# Clean up
+db.destroy_default_schema

lib/langchain/dependency_helper.rb

@@ -15,7 +15,12 @@ class VersionError < ScriptError; end
     # @raise [VersionError] If the gem is installed, but the version does not meet the requirements
     #
     def depends_on(gem_name, req: true)
-      gem(gem_name) # require the gem
+      if gem_name == "sqlite_vec"
+        require "sqlite_vec"
+        return true
+      else
+        gem(gem_name) # require the gem
+      end
 
       return(true) unless defined?(Bundler) # If we're in a non-bundler environment, we're no longer able to determine if we'll meet requirements
 

lib/langchain/llm/ollama.rb

@@ -24,7 +24,7 @@ class Ollama < Base
       llama2: 4_096,
       llama3: 4_096,
       "llama3.1": 4_096,
-      "llama3.2": 4_096,
+      "llama3.2": 3_072,
       llava: 4_096,
       mistral: 4_096,
       "mistral-openorca": 4_096,

lib/langchain/vectorsearch/sqlite_vec.rb

@@ -0,0 +1,154 @@
+# frozen_string_literal: true
+
+require "sqlite_vec"
+module Langchain::Vectorsearch
+  class SqliteVec < Base
+    #
+    # The SQLite vector search adapter using sqlite-vec
+    #
+    # Gem requirements:
+    #     gem "sqlite3", "~> 2.5"
+    #     gem "sqlite_vec", "~> 0.16.0"
+    #
+    # Usage:
+    #     sqlite_vec = Langchain::Vectorsearch::SqliteVec.new(url:, index_name:, llm:, namespace: nil)
+    #
+
+    attr_reader :db, :table_name, :namespace_column, :namespace
+
+    # @param url [String] The path to the SQLite database file (or :memory: for in-memory)
+    # @param index_name [String] The name of the table to use for the index
+    # @param llm [Object] The LLM client to use
+    # @param namespace [String] The namespace to use for the index when inserting/querying
+    def initialize(url:, index_name:, llm:, namespace: nil)
+      depends_on "sqlite3"
+      depends_on "sqlite_vec"
+
+      @db = SQLite3::Database.new(url)
+      @db.enable_load_extension(true)
+      ::SqliteVec.load(@db)
+      @db.enable_load_extension(false)
+
+      @table_name = index_name
+      @namespace_column = "namespace"
+      @namespace = namespace
+
+      super(llm: llm)
+    end
+
+    # Create default schema
+    def create_default_schema
+      @db.execute("CREATE VIRTUAL TABLE IF NOT EXISTS #{table_name} USING vec0(
+        embedding float[#{llm.default_dimensions}],
+        content TEXT,
+        #{namespace_column} TEXT
+      )")
+    end
+
+    # Destroy default schema
+    def destroy_default_schema
+      @db.execute("DROP TABLE IF EXISTS #{table_name}")
+    end
+
+    # Add a list of texts to the index
+    # @param texts [Array<String>] The texts to add to the index
+    # @param ids [Array<String>] The ids to add to the index, in the same order as the texts
+    # @return [Array<Integer>] The ids of the added texts
+    def add_texts(texts:, ids: nil)
+      if ids.nil? || ids.empty?
+        max_rowid = @db.execute("SELECT MAX(rowid) FROM #{table_name}").first.first || 0
+        ids = texts.map.with_index do |_, i|
+          max_rowid + i + 1
+        end
+      end
+
+      @db.transaction do
+        texts.zip(ids).each do |text, id|
+          embedding = llm.embed(text: text).embedding
+          @db.execute(
+            "INSERT INTO #{table_name}(rowid, content, embedding, #{namespace_column}) VALUES (?, ?, ?, ?)",
+            [id, text, embedding.pack("f*"), namespace]
+          )
+        end
+      end
+
+      ids
+    end
+
+    # Update a list of ids and corresponding texts in the index
+    # @param texts [Array<String>] The texts to update in the index
+    # @param ids [Array<String>] The ids to update in the index, in the same order as the texts
+    # @return [Array<Integer>] The ids of the updated texts
+    def update_texts(texts:, ids:)
+      @db.transaction do
+        texts.zip(ids).each do |text, id|
+          embedding = llm.embed(text: text).embedding
+          @db.execute(
+            "UPDATE #{table_name} SET content = ?, embedding = ? WHERE rowid = ?",
+            [text, embedding.pack("f*"), id]
+          )
+        end
+      end
+      ids
+    end
+
+    # Remove a list of texts from the index
+    # @param ids [Array<Integer>] The ids of the texts to remove from the index
+    # @return [Integer] The number of texts removed from the index
+    def remove_texts(ids:)
+      @db.execute("DELETE FROM #{table_name} WHERE rowid IN (#{ids.join(",")})")
+      ids.length
+    end
+
+    # Search for similar texts in the index
+    # @param query [String] The text to search for
+    # @param k [Integer] The number of top results to return
+    # @return [Array<Hash>] The results of the search
+    def similarity_search(query:, k: 4)
+      embedding = llm.embed(text: query).embedding
+      similarity_search_by_vector(embedding: embedding, k: k)
+    end
+
+    # Search for similar texts in the index by vector
+    # @param embedding [Array<Float>] The vector to search for
+    # @param k [Integer] The number of top results to return
+    # @return [Array<Hash>] The results of the search
+    def similarity_search_by_vector(embedding:, k: 4)
+      namespace_condition = namespace ? "AND #{namespace_column} = ?" : ""
+      query_params = [embedding.pack("f*")]
+      query_params << namespace if namespace
+
+      @db.execute(<<-SQL, query_params)
+        SELECT
+          rowid,
+          content,
+          distance
+        FROM #{table_name}
+        WHERE embedding MATCH ?
+        #{namespace_condition}
+        ORDER BY distance
+        LIMIT #{k}
+      SQL
+    end
+
+    # Ask a question and return the answer
+    # @param question [String] The question to ask
+    # @param k [Integer] The number of results to have in context
+    # @yield [String] Stream responses back one String at a time
+    # @return [String] The answer to the question
+    def ask(question:, k: 4, &)
+      search_results = similarity_search(query: question, k: k)
+
+      context = search_results.map { |result| result[1].to_s }
+      context = context.join("\n---\n")
+
+      prompt = generate_rag_prompt(question: question, context: context)
+
+      messages = [{role: "user", content: prompt}]
+      response = llm.chat(messages: messages, &)
+
+      response.context = context
+      response
+    end
+  end
+end