ai_client 0.2.5 → 0.3.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f2b330f59ff7f380306ec354ce5f7197adce8dd101a4b506ab137568f098242a
-  data.tar.gz: f169296b9f20479b1c608f2202d9ab8a8eb416aa80b07f4308ec4648b9c86fcc
+  metadata.gz: 2453f748447f37e755f0087615c845570fe3ea3c4bd06a687947dceedee3e89b
+  data.tar.gz: 79359bcd209448add248514d9a8ee5dc4e0fa69833666df0e779715b574d450a
 SHA512:
-  metadata.gz: ce0e2cddd8ec6935a5bea24125358fbcc8fda902340d7767972142505af41eaeef7d7f6f5791bd7d26469fd45e59020237fd0061e75d6a987d8a4181597cdbe9
-  data.tar.gz: 4b2513fba82802eda13cf26c1af0e73f0903a2140765401ed686540d80480d887f2152e342a5fe3546c89482320f61cc6bd04b4e41b8851668bbda31b2496fc9
+  metadata.gz: 28007804ea1e223b22846cc199c4bd14d7349f5e051bbc7007ced3641e7174c3fac59b7a225bc4926ffc68d5018a35ed96a8809bf959c8007e978bc831770b9a
+  data.tar.gz: 18ce04f9f83d0c12caadab051b81c48f7dbafa73ff0b2a34df28e3ddfb5ff3c088dcec019a15bc2f2dfc909bb6db0a308b512b6711baaab058629705a1448040

data/CHANGELOG.md

@@ -2,8 +2,13 @@
 
 ## Released
 
+### [0.3.0] - 2024-10-13
+- Breaking Change
+- Added new class AiClient::Function to encapsulate the callback functions used as tools in chats.
+- Updated the examples/tools.rb file to use the new function class
+
 ### [0.2.5] - 2024-10-11
-- Added examples/tool.rb to demonstrate use of function callbacks to provide information to the LLM when it needs it.
+- Added examples/tools.rb to demonstrate use of function callbacks to provide information to the LLM when it needs it.
 
 ### [0.2.4] - 2024-10-10
 - constrained gem omniai-openai to version 1.8.3+ for access to open_router.ai

data/README.md

@@ -2,6 +2,8 @@
 
 First and foremost a big **THANK YOU** to [Kevin Sylvestre](https://ksylvest.com/) for his gem [OmniAI](https://github.com/ksylvest/omniai) and [Olympia](https://olympia.chat/) for their [open_router gem](https://github.com/OlympiaAI/open_router) upon which this effort depends.
 
+Version 0.3.0 has a breaking change w/r/t how [Callback Functions (aka Tools)](#callback-functions-aka-tools) are defined and used.
+
 See the  [change log](CHANGELOG.md) for recent modifications.
 
 
@@ -33,7 +35,8 @@ See the  [change log](CHANGELOG.md) for recent modifications.
         - [transcribe](#transcribe)
     - [Options](#options)
     - [Advanced Prompts](#advanced-prompts)
-    - [Advanced Prompts with Tools](#advanced-prompts-with-tools)
+    - [Callback Functions (aka Tools)](#callback-functions-aka-tools)
+        - [Defining a Callback Function](#defining-a-callback-function)
   - [Best ?? Practices](#best--practices)
   - [OmniAI and OpenRouter](#omniai-and-openrouter)
   - [Contributing](#contributing)
@@ -278,13 +281,55 @@ completion #=> 'The photos are of a cat, a dog, and a hamster.'
 
 Of course if `client.config.return_raw` is true, the completion value will be the complete response object.
 
-### Advanced Prompts with Tools
 
-One of the latest innovations in LLMs is the ability to use functions (aka tools) as `callbacks` to gather more information or to execute a task at the direction of the LLM prompt processing.
+### Callback Functions (aka Tools)
+
+With the release of version 0.3.0, the way callback functions (also referred to as tools) are defined in the `ai_client` gem has undergone significant changes. This section outlines the new approach in detail.  These changes are designed to create a clearer and more robust interface for developers when working with callback functions. If you encounter any issues while updating your functions, please consult the official documentation or raise an issue in the repository.
+
+##### Defining a Callback Function
+
+To define a callback function, you need to create a subclass of `AiClient::Function`. In this subclass, both the `call` and `details` methods must be implemented.
+
+**Example**
+
+Here's an example illustrating how to define a callback function using the new convention:
+
+```ruby
+class WeatherFunction < AiClient::Function
+  # The call class method returns a String to be used by the LLM
+  def self.call(location:, unit: 'Celsius')
+    "#{rand(20..50)}° #{unit} in #{location}"
+  end
+
+  # The details method must return a hash with metadata about the function.
+  def self.details
+    {
+      name:         'weather',
+      description:  "Lookup the weather in a location",
+      parameters:   AiClient::Tool::Parameters.new(
+        properties: {
+          location: AiClient::Tool::Property.string(description: 'e.g. Toronto'),
+          unit:     AiClient::Tool::Property.string(enum: %w[Celsius Fahrenheit]),
+        },
+        required: [:location]
+      )
+    }
+  end
+end
+
+# Register the WeatherFunction for use.
+WeatherFunction.register
+
+# Use the *.details[:name] value to reference the tools available for
+# the LLM to use in processing the prompt.
+response = AI.chat("what is the weather in London", tools: ['weather'])
+```
 
-See [blog post](https://ksylvest.com/posts/2024-08-16/using-omniai-to-leverage-tools-with-llms) by Kevin Sylvestre, author of the OmniAI gem.
+In this example:
+- The `call` method is defined to accept named parameters: `location` and `unit`. The default value for `unit` is set to `'Celsius'`.
+- The `details` method provides metadata about the function, ensuring that the parameters section clearly indicates which parameters are required.
 
-Take a look at the [examples/tools.rb](examples/tools.rb) file to see different ways in which these callable processes can be defined.
+See the [examples/tools.rb file](examples/tools.rb) for additional examples.
 
 
 ## Best ?? Practices
@@ -302,6 +347,7 @@ AI.speak "warning  Will Robinson! #{bad_things_happened}"
 
 Using the constant for the instance allows you to reference the same client instance inside any method through out your application.  Of course it does not apply to only one instance.  You could assign multiple instances for different models/providers.  For example you could have `AI` for your primary client and `AIbackup` for a fallback client in case you have a problem on the primary; or, maybe `Vectorizer` as a client name tied to a model specializing in embedding vectorization.
 
+
 ## OmniAI and OpenRouter
 
 Both OmniAI and OpenRouter have similar goals - to provide a common interface to multiple providers and LLMs.  OmniAI is a Ruby gem that supports specific providers directly using a common-ish API.  You incur costs directly from those providers for which you have individual API keys (aka access tokens.) OpenRouter, on the other hand, is a web service that also establishes a common API for many providers and models; however, OpenRouter adds a small fee on top of the fee charged by those providers.  You trade off cost for flexibility.  With OpenRouter you only need one API key (OPEN_ROUTER_API_KEY) to access all of its supported services.

data/examples/tools.rb

@@ -1,90 +1,123 @@
 #!/usr/bin/env ruby
 # examples/tools.rb
-# See: https://ksylvest.com/posts/2024-08-16/using-omniai-to-leverage-tools-with-llms
+#
+# Uses the AiClient::Function class to encapsulate the
+# tools used as callback functions when specified in a
+# chat prompt.
+
 
 require_relative 'common'
 
 AI = AiClient.new('gpt-4o')
 
-box "omniai-openai's random temp example"
+box "Random Weather (temperature) Example"
+title "Uses two named parameters to the callback function"
+
+# Example subclass implementation
+class WeatherFunction < AiClient::Function
+  def self.call(location:, unit: 'Celsius')
+    "#{rand(20..50)}° #{unit} in #{location}"
+  end
 
-my_weather_function = Proc.new do |location:, unit: 'Celsius'| 
-  "#{rand(20..50)}° #{unit} in #{location}"
+  # Encapsulates registration details for the function
+  def self.details
+    # SMELL:  reconcile regester_tool and details
+    {
+      name:         'weather', # Must be a String
+      description:  "Lookup the weather in a location",
+      parameters:   AiClient::Tool::Parameters.new(
+        properties: {
+          location: AiClient::Tool::Property.string(description: 'e.g. Toronto'),
+          unit:     AiClient::Tool::Property.string(enum: %w[Celsius Fahrenheit]),
+        },
+        required: [:location]
+      )
+    }
+  end
 end
 
-weather = AiClient::Tool.new(
-  my_weather_function,
-  name: 'weather',
-  description: 'Lookup the weather in a location',
-  parameters: AiClient::Tool::Parameters.new(
-    properties: {
-      location: AiClient::Tool::Property.string(description: 'e.g. Toronto'),
-      unit: AiClient::Tool::Property.string(enum: %w[Celsius Fahrenheit]),
-    },
-    required: %i[location]
-  )
-)
+# Register the tool for MyFunction
+WeatherFunction.register
 
 simple_prompt = <<~TEXT
   What is the weather in "London" in Celsius and "Paris" in Fahrenheit?
   Also what are some ideas for activities in both cities given the weather?
 TEXT
 
-response = AI.chat(simple_prompt, tools: [weather])
+response = AI.chat(
+                    simple_prompt, 
+                    tools: ['weather'] # must match the details[:name] value
+                  )
+
 puts response
 
+
 ##########################################
 box "Accessing a database to get information"
+title "Uses one named parameter to the callback function"
 
-llm_db_function = Proc.new do |params|
-  records = AiClient::LLM.where(id: /#{params[:model_name]}/i)
-  records.inspect
+class LLMDetailedFunction < AiClient::Function
+  def self.call(model_name:)
+    records = AiClient::LLM.where(id: /#{model_name}/i)&.first
+    records.inspect
+  end
+
+  def self.details
+    {
+      name:         'llm_db',
+      description:  'lookup details about an LLM model name',
+      parameters:   AiClient::Tool::Parameters.new(
+        properties: {
+          model_name: AiClient::Tool::Property.string
+        },
+        required: %i[model_name]
+      )
+    }  
+  end
 end
 
+# Registering the LLM detail function
+LLMDetailedFunction.register
 
-llm_db = AiClient::Tool.new(
-  llm_db_function,
-  name: 'llm_db',
-  description: 'lookup details about an LLM model name',
-  parameters: AiClient::Tool::Parameters.new(
-    properties: {
-      model_name: AiClient::Tool::Property.string
-    },
-    required: %i[model_name]
-  )
-)
+simple_prompt = <<~PROMPT
+  Get the details on an LLM model named 'bison' from the models database 
+  of #{AiClient::LLM.count} models. Format the details for the model
+  using markdown.  Format pricing information in terms of number of
+  tokens per US dollar.
+PROMPT
 
-response = AI.chat("Get details on an LLM model named bison.  Which one is the cheapest per prompt token.", tools: [llm_db])
+response = AI.chat(simple_prompt, tools: ['llm_db'])
 puts response
 
-##########################################
 
-# TODO: Look at creating a better function
-#       process such that the tools parameter
-#       is an Array of Symbols which is
-#       maintained as a class variable.
-#       The symboles are looked up and the
-#       proper instance is inserted in its
-#       place.
+
+##########################################
 
 box "Using a function class and multiple tools"
+title "Callback function has no parameters; but uses two functions"
 
-class FunctionClass
+class PerfectDateFunction < AiClient::Function
   def self.call
-    "April 25th its not to hot nor too cold."
+    "April 25th, it's not too hot nor too cold."
   end
 
-  def function(my_name)
-    AiClient::Tool.new(
-      self.class, # with a self.call method
-      name: my_name,
-      description: 'what is the perfect date'
-    )
+  def self.details
+    {
+      name:         'perfect_date',
+      description:  'what is the perfect date'
+    }
   end
 end
 
-perfect_date = FunctionClass.new.function('perfect_date')
+# Registering the perfect date function
+PerfectDateFunction.register
 
-response = AI.chat("what is the perfect date for paris weather?", tools: [weather, perfect_date])
+response = AI.chat("what is the perfect date for current weather in Paris?", 
+                   tools: %w[weather perfect_date])
 puts response
 puts
+
+debug_me{[
+  #'AiClient::Function.registry',
+  'AiClient::Function.functions'
+]}

data/lib/ai_client/chat.rb

@@ -10,7 +10,19 @@ class AiClient
   #   tools:        @tools    [Array<OmniAI::Tool>] optional
   #   temperature:  @temperature  [Float, nil] optional
 
-  def chat(messages, **params)
+  def chat(messages, **params)    
+    if params.has_key? :tools
+      tools = params[:tools]
+      if tools.is_a? Array
+        tools.map!{|function_name| AiClient::Function.registry[function_name]}
+      elsif true == tools
+        tools = AiClient::Function.registry.values
+      else
+        raise 'what is this'
+      end
+      params[:tools] = tools
+    end
+
     result = call_with_middlewares(:chat_without_middlewares, messages, **params)
     @last_response = result
     raw? ? result : content

data/lib/ai_client/function.rb

@@ -0,0 +1,100 @@
+# lib/ai_client/function.rb
+
+class AiClient
+
+  # The Function class serves as a base class for creating functions
+  # that can be registered and managed within the AiClient.
+  #
+  # Subclasses must implement the `call` and `details` methods
+  # to define their specific behavior and properties.
+  #
+  class Function
+    @@registry = {} # key is known by name (from details) and value is the AiClient::Tool
+
+    class << self
+
+      # Calls the function with the provided parameters.
+      #
+      # @param params [Hash] Named parameters required by the function.
+      # @raise [NotImplementedError] if not implemented in the subclass.
+      #
+      def call(**params)
+        raise NotImplementedError, "You must implement the call method"
+      end
+
+      # Provides the details about the function including its metadata.
+      #
+      # @return [Hash] Metadata containing details about the function.
+      # @raise [NotImplementedError] if not implemented in the subclass.
+      #
+      def details
+        raise NotImplementedError, "You must implement the details method"
+      end
+
+
+      # Registers a tool with the specified properties and parameters.
+      #
+      # This method creates an instance of AiClient::Tool with the
+      # function class and its details and adds it to the registry.
+      #
+      def register
+        this_tool = AiClient::Tool.new(
+                      self, # This is the sub-class
+                      **details
+                    )
+
+        registry[known_by] = this_tool
+      end
+      alias_method :enable, :register
+
+
+      # Disables the function by removing its name from the registry.
+      #
+      # @return [void]
+      #
+      def disable
+        registry.delete(known_by)
+      end
+
+
+      # Returns a list of enabled functions.
+      #
+      # @return [Array<Symbol>] Sorted list of function names.
+      #
+      def functions
+        registry.keys.sort
+      end
+
+
+      # Returns the registry of currently registered functions.
+      # This method is private to limit access to the registry's state.
+      #
+      # @return [Hash] The registry of registered functions.
+      #
+      def registry
+        @@registry
+      end
+
+
+      # Returns the name under which the function is known.
+      #
+      # @return [Symbol] The symbol representation of the function's name.
+      #
+      def known_by
+        details[:name]
+      end
+
+
+      private
+
+      # Converts the class name to a symbol (e.g., MyFunction -> :my_function).
+      #
+      # @return [Symbol] The function name derived from the class name.
+      #
+      def function_name
+        name.split('::').last.gsub(/([a-z])([A-Z])/, '\1_\2').downcase
+      end
+    end
+  end
+end
+

data/lib/ai_client/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 class AiClient
-  VERSION = "0.2.5"
+  VERSION = "0.3.0"
 end

data/lib/ai_client.rb

@@ -29,6 +29,7 @@ require_relative 'ai_client/middleware'
 require_relative 'ai_client/open_router_extensions'
 require_relative 'ai_client/llm' # SMELL: must come after the open router stuff
 require_relative 'ai_client/tool'
+require_relative 'ai_client/function'
 
 # Create a generic client instance using only model name
 #   client = AiClient.new('gpt-3.5-turbo')

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: ai_client
 version: !ruby/object:Gem::Version
-  version: 0.2.5
+  version: 0.3.0
 platform: ruby
 authors:
 - Dewayne VanHoozer
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2024-10-11 00:00:00.000000000 Z
+date: 2024-10-14 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: active_hash
@@ -226,6 +226,7 @@ files:
 - lib/ai_client/config.yml
 - lib/ai_client/configuration.rb
 - lib/ai_client/embed.rb
+- lib/ai_client/function.rb
 - lib/ai_client/llm.rb
 - lib/ai_client/logger_middleware.rb
 - lib/ai_client/middleware.rb