raif 1.0.0 → 1.1.0

data/lib/raif/llm_registry.rb

@@ -0,0 +1,205 @@
+# frozen_string_literal: true
+
+module Raif
+  class << self
+    attr_accessor :llm_registry
+  end
+
+  def self.register_llm(llm_class, llm_config)
+    llm = llm_class.new(**llm_config)
+
+    unless llm.valid?
+      raise ArgumentError, "The LLM you tried to register is invalid: #{llm.errors.full_messages.join(", ")}"
+    end
+
+    @llm_registry ||= {}
+    @llm_registry[llm.key] = llm_config.merge(llm_class: llm_class)
+  end
+
+  def self.llm(model_key)
+    llm_config = llm_registry[model_key]
+
+    if llm_config.nil?
+      raise ArgumentError, "No LLM found for model key: #{model_key}. Available models: #{available_llm_keys.join(", ")}"
+    end
+
+    llm_class = llm_config[:llm_class]
+    llm_class.new(**llm_config.except(:llm_class))
+  end
+
+  def self.available_llms
+    llm_registry.values
+  end
+
+  def self.available_llm_keys
+    llm_registry.keys
+  end
+
+  def self.llm_config(model_key)
+    llm_registry[model_key]
+  end
+
+  def self.default_llms
+    {
+      Raif::Llms::OpenAi => [
+        {
+          key: :open_ai_gpt_4o_mini,
+          api_name: "gpt-4o-mini",
+          input_token_cost: 0.15 / 1_000_000,
+          output_token_cost: 0.6 / 1_000_000,
+        },
+        {
+          key: :open_ai_gpt_4o,
+          api_name: "gpt-4o",
+          input_token_cost: 2.5 / 1_000_000,
+          output_token_cost: 10.0 / 1_000_000,
+        },
+        {
+          key: :open_ai_gpt_3_5_turbo,
+          api_name: "gpt-3.5-turbo",
+          input_token_cost: 0.5 / 1_000_000,
+          output_token_cost: 1.5 / 1_000_000,
+          model_provider_settings: { supports_structured_outputs: false }
+        },
+        {
+          key: :open_ai_gpt_4_1,
+          api_name: "gpt-4.1",
+          input_token_cost: 2.0 / 1_000_000,
+          output_token_cost: 8.0 / 1_000_000,
+        },
+        {
+          key: :open_ai_gpt_4_1_mini,
+          api_name: "gpt-4.1-mini",
+          input_token_cost: 0.4 / 1_000_000,
+          output_token_cost: 1.6 / 1_000_000,
+        },
+        {
+          key: :open_ai_gpt_4_1_nano,
+          api_name: "gpt-4.1-nano",
+          input_token_cost: 0.1 / 1_000_000,
+          output_token_cost: 0.4 / 1_000_000,
+        },
+      ],
+      Raif::Llms::Anthropic => [
+        {
+          key: :anthropic_claude_4_sonnet,
+          api_name: "claude-sonnet-4-20250514",
+          input_token_cost: 3.0 / 1_000_000,
+          output_token_cost: 15.0 / 1_000_000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :anthropic_claude_4_opus,
+          api_name: "claude-opus-4-20250514",
+          input_token_cost: 15.0 / 1_000_000,
+          output_token_cost: 75.0 / 1_000_000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :anthropic_claude_3_7_sonnet,
+          api_name: "claude-3-7-sonnet-latest",
+          input_token_cost: 3.0 / 1_000_000,
+          output_token_cost: 15.0 / 1_000_000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :anthropic_claude_3_5_sonnet,
+          api_name: "claude-3-5-sonnet-latest",
+          input_token_cost: 3.0 / 1_000_000,
+          output_token_cost: 15.0 / 1_000_000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :anthropic_claude_3_5_haiku,
+          api_name: "claude-3-5-haiku-latest",
+          input_token_cost: 0.8 / 1_000_000,
+          output_token_cost: 4.0 / 1_000_000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :anthropic_claude_3_opus,
+          api_name: "claude-3-opus-latest",
+          input_token_cost: 15.0 / 1_000_000,
+          output_token_cost: 75.0 / 1_000_000,
+          max_completion_tokens: 4096
+        },
+      ],
+      Raif::Llms::BedrockClaude => [
+        {
+          key: :bedrock_claude_4_sonnet,
+          api_name: "anthropic.claude-sonnet-4-20250514-v1:0",
+          input_token_cost: 0.003 / 1000,
+          output_token_cost: 0.015 / 1000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :bedrock_claude_4_opus,
+          api_name: "anthropic.claude-opus-4-20250514-v1:0",
+          input_token_cost: 0.015 / 1000,
+          output_token_cost: 0.075 / 1000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :bedrock_claude_3_5_sonnet,
+          api_name: "anthropic.claude-3-5-sonnet-20241022-v2:0",
+          input_token_cost: 0.003 / 1000,
+          output_token_cost: 0.015 / 1000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :bedrock_claude_3_7_sonnet,
+          api_name: "anthropic.claude-3-7-sonnet-20250219-v1:0",
+          input_token_cost: 0.003 / 1000,
+          output_token_cost: 0.015 / 1000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :bedrock_claude_3_5_haiku,
+          api_name: "anthropic.claude-3-5-haiku-20241022-v1:0",
+          input_token_cost: 0.0008 / 1000,
+          output_token_cost: 0.004 / 1000,
+          max_completion_tokens: 8192
+        },
+        {
+          key: :bedrock_claude_3_opus,
+          api_name: "anthropic.claude-3-opus-20240229-v1:0",
+          input_token_cost: 0.015 / 1000,
+          output_token_cost: 0.075 / 1000,
+          max_completion_tokens: 4096
+        },
+      ],
+      Raif::Llms::OpenRouter => [
+        {
+          key: :open_router_claude_3_7_sonnet,
+          api_name: "anthropic/claude-3.7-sonnet",
+          input_token_cost: 3.0 / 1_000_000,
+          output_token_cost: 15.0 / 1_000_000,
+        },
+        {
+          key: :open_router_llama_3_3_70b_instruct,
+          api_name: "meta-llama/llama-3.3-70b-instruct",
+          input_token_cost: 0.10 / 1_000_000,
+          output_token_cost: 0.25 / 1_000_000,
+        },
+        {
+          key: :open_router_llama_3_1_8b_instruct,
+          api_name: "meta-llama/llama-3.1-8b-instruct",
+          input_token_cost: 0.02 / 1_000_000,
+          output_token_cost: 0.03 / 1_000_000,
+        },
+        {
+          key: :open_router_gemini_2_0_flash,
+          api_name: "google/gemini-2.0-flash-001",
+          input_token_cost: 0.1 / 1_000_000,
+          output_token_cost: 0.4 / 1_000_000,
+        },
+        {
+          key: :open_router_deepseek_chat_v3,
+          api_name: "deepseek/deepseek-chat-v3-0324",
+          input_token_cost: 0.27 / 1_000_000,
+          output_token_cost: 1.1 / 1_000_000,
+        },
+      ]
+    }
+  end
+end

data/lib/raif/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Raif
-  VERSION = "1.0.0"
+  VERSION = "1.1.0"
 end

data/lib/raif.rb

@@ -6,8 +6,12 @@ require "raif/engine"
 require "raif/configuration"
 require "raif/errors"
 require "raif/utils"
-require "raif/default_llms"
+require "raif/llm_registry"
+require "raif/embedding_model_registry"
+require "raif/json_schema_builder"
+
 require "faraday"
+require "json-schema"
 require "loofah"
 require "pagy"
 require "reverse_markdown"
@@ -16,7 +20,6 @@ require "turbo-rails"
 module Raif
   class << self
     attr_accessor :configuration
-    attr_accessor :llm_registry
 
     attr_writer :logger
   end
@@ -32,34 +35,4 @@ module Raif
   def self.logger
     @logger ||= Rails.logger
   end
-
-  def self.register_llm(llm_class, llm_config)
-    llm = llm_class.new(**llm_config)
-
-    unless llm.valid?
-      raise ArgumentError, "The LLM you tried to register is invalid: #{llm.errors.full_messages.join(", ")}"
-    end
-
-    @llm_registry ||= {}
-    @llm_registry[llm.key] = llm_config.merge(llm_class: llm_class)
-  end
-
-  def self.llm(model_key)
-    llm_config = llm_registry[model_key]
-
-    if llm_config.nil?
-      raise ArgumentError, "No LLM found for model key: #{model_key}. Available models: #{available_llm_keys.join(", ")}"
-    end
-
-    llm_class = llm_config[:llm_class]
-    llm_class.new(**llm_config.except(:llm_class))
-  end
-
-  def self.available_llms
-    llm_registry.values
-  end
-
-  def self.available_llm_keys
-    llm_registry.keys
-  end
 end

data/lib/tasks/raif_tasks.rake

@@ -1,6 +1,11 @@
 # frozen_string_literal: true
 
-# desc "Explaining what the task does"
-# task :raif do
-#   # Task goes here
-# end
+namespace :raif do
+  namespace :install do
+    desc "Copy migrations from Raif to host application"
+    task :migrations do
+      ENV["FROM"] = "raif"
+      Rake::Task["railties:install:migrations"].invoke
+    end
+  end
+end

metadata

@@ -1,29 +1,15 @@
 --- !ruby/object:Gem::Specification
 name: raif
 version: !ruby/object:Gem::Version
-  version: 1.0.0
+  version: 1.1.0
 platform: ruby
 authors:
 - Ben Roesch
 - Brian Leslie
 bindir: bin
 cert_chain: []
-date: 2025-04-02 00:00:00.000000000 Z
+date: 2025-05-23 00:00:00.000000000 Z
 dependencies:
-- !ruby/object:Gem::Dependency
-  name: aws-sdk-bedrock
-  requirement: !ruby/object:Gem::Requirement
-    requirements:
-    - - ">="
-      - !ruby/object:Gem::Version
-        version: '0'
-  type: :runtime
-  prerelease: false
-  version_requirements: !ruby/object:Gem::Requirement
-    requirements:
-    - - ">="
-      - !ruby/object:Gem::Version
-        version: '0'
 - !ruby/object:Gem::Dependency
   name: aws-sdk-bedrockruntime
   requirement: !ruby/object:Gem::Requirement
@@ -157,6 +143,7 @@ files:
 - app/assets/javascript/raif/stream_actions/raif_scroll_to_bottom.js
 - app/assets/stylesheets/raif.scss
 - app/assets/stylesheets/raif/admin/conversation.scss
+- app/assets/stylesheets/raif/admin/stats.scss
 - app/assets/stylesheets/raif/loader.scss
 - app/assets/stylesheets/raif_admin.scss
 - app/controllers/raif/admin/agents_controller.rb
@@ -164,6 +151,8 @@ files:
 - app/controllers/raif/admin/conversations_controller.rb
 - app/controllers/raif/admin/model_completions_controller.rb
 - app/controllers/raif/admin/model_tool_invocations_controller.rb
+- app/controllers/raif/admin/stats/tasks_controller.rb
+- app/controllers/raif/admin/stats_controller.rb
 - app/controllers/raif/admin/tasks_controller.rb
 - app/controllers/raif/application_controller.rb
 - app/controllers/raif/conversation_entries_controller.rb
@@ -182,14 +171,26 @@ files:
 - app/models/raif/concerns/has_llm.rb
 - app/models/raif/concerns/has_requested_language.rb
 - app/models/raif/concerns/invokes_model_tools.rb
+- app/models/raif/concerns/json_schema_definition.rb
 - app/models/raif/concerns/llm_response_parsing.rb
+- app/models/raif/concerns/llm_temperature.rb
+- app/models/raif/concerns/llms/anthropic/message_formatting.rb
+- app/models/raif/concerns/llms/bedrock_claude/message_formatting.rb
+- app/models/raif/concerns/llms/message_formatting.rb
+- app/models/raif/concerns/llms/open_ai/message_formatting.rb
 - app/models/raif/conversation.rb
 - app/models/raif/conversation_entry.rb
+- app/models/raif/embedding_model.rb
+- app/models/raif/embedding_models/bedrock_titan.rb
+- app/models/raif/embedding_models/open_ai.rb
 - app/models/raif/llm.rb
 - app/models/raif/llms/anthropic.rb
 - app/models/raif/llms/bedrock_claude.rb
 - app/models/raif/llms/open_ai.rb
+- app/models/raif/llms/open_router.rb
 - app/models/raif/model_completion.rb
+- app/models/raif/model_file_input.rb
+- app/models/raif/model_image_input.rb
 - app/models/raif/model_tool.rb
 - app/models/raif/model_tool_invocation.rb
 - app/models/raif/model_tools/agent_final_answer.rb
@@ -212,6 +213,8 @@ files:
 - app/views/raif/admin/model_tool_invocations/_model_tool_invocation.html.erb
 - app/views/raif/admin/model_tool_invocations/index.html.erb
 - app/views/raif/admin/model_tool_invocations/show.html.erb
+- app/views/raif/admin/stats/index.html.erb
+- app/views/raif/admin/stats/tasks/index.html.erb
 - app/views/raif/admin/tasks/_task.html.erb
 - app/views/raif/admin/tasks/index.html.erb
 - app/views/raif/admin/tasks/show.html.erb
@@ -234,9 +237,16 @@ files:
 - config/locales/en.yml
 - config/routes.rb
 - db/migrate/20250224234252_create_raif_tables.rb
+- db/migrate/20250421202149_add_response_format_to_raif_conversations.rb
+- db/migrate/20250424200755_add_cost_columns_to_raif_model_completions.rb
+- db/migrate/20250424232946_add_created_at_indexes.rb
+- db/migrate/20250502155330_add_status_indexes_to_raif_tasks.rb
+- db/migrate/20250507155314_add_retry_count_to_raif_model_completions.rb
 - lib/generators/raif/agent/agent_generator.rb
 - lib/generators/raif/agent/templates/agent.rb.tt
+- lib/generators/raif/agent/templates/application_agent.rb.tt
 - lib/generators/raif/conversation/conversation_generator.rb
+- lib/generators/raif/conversation/templates/application_conversation.rb.tt
 - lib/generators/raif/conversation/templates/conversation.rb.tt
 - lib/generators/raif/install/install_generator.rb
 - lib/generators/raif/install/templates/initializer.rb
@@ -248,17 +258,20 @@ files:
 - lib/generators/raif/views_generator.rb
 - lib/raif.rb
 - lib/raif/configuration.rb
-- lib/raif/default_llms.rb
+- lib/raif/embedding_model_registry.rb
 - lib/raif/engine.rb
 - lib/raif/errors.rb
 - lib/raif/errors/action_not_authorized_error.rb
-- lib/raif/errors/anthropic/api_error.rb
 - lib/raif/errors/invalid_config_error.rb
 - lib/raif/errors/invalid_conversation_type_error.rb
+- lib/raif/errors/invalid_model_file_input_error.rb
+- lib/raif/errors/invalid_model_image_input_error.rb
 - lib/raif/errors/invalid_user_tool_type_error.rb
-- lib/raif/errors/open_ai/api_error.rb
 - lib/raif/errors/open_ai/json_schema_error.rb
+- lib/raif/errors/unsupported_feature_error.rb
+- lib/raif/json_schema_builder.rb
 - lib/raif/languages.rb
+- lib/raif/llm_registry.rb
 - lib/raif/rspec.rb
 - lib/raif/utils.rb
 - lib/raif/utils/html_to_markdown_converter.rb

data/lib/raif/default_llms.rb

@@ -1,37 +0,0 @@
-# frozen_string_literal: true
-
-module Raif
-  def self.default_llms
-    {
-      Raif::Llms::OpenAi => [
-        {
-          key: :open_ai_gpt_4o_mini,
-          api_name: "gpt-4o-mini",
-          model_provider_settings: { supports_structured_outputs: true }
-        },
-        {
-          key: :open_ai_gpt_4o,
-          api_name: "gpt-4o",
-          model_provider_settings: { supports_structured_outputs: true }
-        },
-        {
-          key: :open_ai_gpt_3_5_turbo,
-          api_name: "gpt-3.5-turbo",
-          model_provider_settings: { supports_structured_outputs: false }
-        },
-      ],
-      Raif::Llms::Anthropic => [
-        { key: :anthropic_claude_3_7_sonnet, api_name: "claude-3-7-sonnet-latest", max_completion_tokens: 8192 },
-        { key: :anthropic_claude_3_5_sonnet, api_name: "claude-3-5-sonnet-latest", max_completion_tokens: 8192 },
-        { key: :anthropic_claude_3_5_haiku, api_name: "claude-3-5-haiku-latest", max_completion_tokens: 8192 },
-        { key: :anthropic_claude_3_opus, api_name: "claude-3-opus-latest", max_completion_tokens: 4096 },
-      ],
-      Raif::Llms::BedrockClaude => [
-        { key: :bedrock_claude_3_5_sonnet, api_name: "anthropic.claude-3-5-sonnet-20241022-v2:0", max_completion_tokens: 8192 },
-        { key: :bedrock_claude_3_7_sonnet, api_name: "anthropic.claude-3-7-sonnet-20250219-v1:0", max_completion_tokens: 8192 },
-        { key: :bedrock_claude_3_5_haiku, api_name: "anthropic.claude-3-5-haiku-20241022-v1:0", max_completion_tokens: 8192 },
-        { key: :bedrock_claude_3_opus, api_name: "anthropic.claude-3-opus-20240229-v1:0", max_completion_tokens: 4096 },
-      ]
-    }
-  end
-end