Add Ollama as a supported provider

docs/guides/getting-started.md

@@ -33,6 +33,16 @@ RubyLLM.configure do |config|
 end
 ```
 
+Alternatively, you can point to an Ollama instance; here it is configured to the most common address when Ollama is installed locally:
+```
+RubyLLM.configure do |config|
+  config.ollama_api_base_url = 'http://localhost:11434'
+end
+
+# needs to be called to populate Ollama models before using any
+RubyLLM.models.refresh!
+```
+
 ## Your First Chat
 
 Let's start with a simple chat interaction:
@@ -64,6 +74,10 @@ claude_chat.ask "Tell me about Ruby programming language"
 # Use Gemini
 gemini_chat = RubyLLM.chat(model: 'gemini-2.0-flash')
 gemini_chat.ask "What are the best Ruby gems for machine learning?"
+
+# Use an Ollama model:
+ollama_chat = RubyLLM.chat(model: 'gemma3:latest')
+ollama_chat.ask "What is Alphabet?"
 ```
 
 ## Exploring Available Models
@@ -161,4 +175,4 @@ Now that you've got the basics down, you're ready to explore more advanced featu
 
 - [Chatting with AI]({% link guides/chat.md %}) - Learn more about chat capabilities
 - [Using Tools]({% link guides/tools.md %}) - Let AI use your Ruby code
-- [Rails Integration]({% link guides/rails.md %}) - Persist chats in your Rails apps
+- [Rails Integration]({% link guides/rails.md %}) - Persist chats in your Rails apps

lib/ruby_llm.rb

@@ -15,7 +15,8 @@
   'llm' => 'LLM',
   'openai' => 'OpenAI',
   'api' => 'API',
-  'deepseek' => 'DeepSeek'
+  'deepseek' => 'DeepSeek',
+  'ollama' => 'Ollama'
 )
 loader.setup
 
@@ -68,6 +69,7 @@ def logger
 RubyLLM::Provider.register :anthropic, RubyLLM::Providers::Anthropic
 RubyLLM::Provider.register :gemini, RubyLLM::Providers::Gemini
 RubyLLM::Provider.register :deepseek, RubyLLM::Providers::DeepSeek
+RubyLLM::Provider.register :ollama, RubyLLM::Providers::Ollama
 
 if defined?(Rails::Railtie)
   require 'ruby_llm/railtie'

lib/ruby_llm/configuration.rb

@@ -14,6 +14,7 @@ class Configuration
                   :anthropic_api_key,
                   :gemini_api_key,
                   :deepseek_api_key,
+                  :ollama_api_base_url,
                   :default_model,
                   :default_embedding_model,
                   :default_image_model,

lib/ruby_llm/provider.rb

@@ -19,6 +19,8 @@ def complete(messages, tools:, temperature:, model:, &block)
       end
 
       def list_models
+        return [] unless enabled?
+
         response = connection.get(models_url) do |req|
           req.headers.merge! headers
         end
@@ -123,11 +125,21 @@ def to_json_stream(&block) # rubocop:disable Metrics/MethodLength
               RubyLLM.logger.debug "Accumulating error chunk: #{chunk}"
             end
           else
-            parser.feed(chunk) do |_type, data|
-              unless data == '[DONE]'
-                parsed_data = JSON.parse(data)
-                block.call(parsed_data)
+            content_type = env.response_headers['content-type']
+
+            case content_type
+            when 'text/event-stream'
+              parser.feed(chunk) do |_type, data|
+                unless data == '[DONE]'
+                  parsed_data = JSON.parse(data)
+                  block.call(parsed_data)
+                end
               end
+            when 'application/x-ndjson'
+              parsed_data = JSON.parse(chunk)
+              block.call(parsed_data)
+            else
+              raise NotImplementedError, "unsupported content-type in streaming response: #{content_type}"
             end
           end
         end

lib/ruby_llm/providers/anthropic.rb

@@ -15,6 +15,10 @@ module Anthropic
 
       module_function
 
+      def enabled?
+        !!RubyLLM.config.anthropic_api_key
+      end
+
       def api_base
         'https://api.anthropic.com'
       end

lib/ruby_llm/providers/deepseek.rb

@@ -8,6 +8,10 @@ module DeepSeek
 
       module_function
 
+      def enabled?
+        !!RubyLLM.config.deepseek_api_key
+      end
+
       def api_base
         'https://api.deepseek.com'
       end

lib/ruby_llm/providers/gemini.rb

@@ -15,6 +15,10 @@ module Gemini
 
       module_function
 
+      def enabled?
+        !!RubyLLM.config.gemini_api_key
+      end
+
       def api_base
         'https://generativelanguage.googleapis.com/v1beta'
       end

lib/ruby_llm/providers/gemini/models.rb

@@ -11,6 +11,8 @@ def models_url
         end
 
         def list_models
+          return [] unless enabled?
+
           response = connection.get("models?key=#{RubyLLM.config.gemini_api_key}") do |req|
             req.headers.merge! headers
           end

lib/ruby_llm/providers/ollama.rb

@@ -0,0 +1,36 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Providers
+    # Native Ollama API implementation
+    module Ollama
+      extend Provider
+      extend Ollama::Chat
+      extend Ollama::Embeddings
+      extend Ollama::Models
+      extend Ollama::Streaming
+
+      module_function
+
+      def enabled?
+        !!RubyLLM.config.ollama_api_base_url
+      end
+
+      def api_base
+        RubyLLM.config.ollama_api_base_url
+      end
+
+      def headers
+        {}
+      end
+
+      def capabilities
+        Ollama::Capabilities
+      end
+
+      def slug
+        'ollama'
+      end
+    end
+  end
+end

lib/ruby_llm/providers/ollama/capabilities.rb

@@ -0,0 +1,138 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Providers
+    module Ollama
+      # Determines capabilities for Ollama
+      module Capabilities # rubocop:disable Metrics/ModuleLength
+        module_function
+
+        # Returns the context window size (input token limit) for the given model
+        # @param model_id [String] the model identifier
+        # @return [Integer] the context window size in tokens
+        def context_window_for(model_id)
+          # FIXME: revise
+          4_192 # Sensible (and conservative) default for unknown models
+        end
+
+        # Returns the maximum output tokens for the given model
+        # @param model_id [String] the model identifier
+        # @return [Integer] the maximum output tokens
+        def max_tokens_for(model_id)
+          # FIXME: revise
+          32_768
+        end
+
+        # Returns the input price per million tokens for the given model
+        # @param model_id [String] the model identifier
+        # @return [Float] the price per million tokens in USD
+        def input_price_for(model_id)
+          0.0
+        end
+
+        # Returns the output price per million tokens for the given model
+        # @param model_id [String] the model identifier
+        # @return [Float] the price per million tokens in USD
+        def output_price_for(model_id)
+          0.0
+        end
+
+        # Determines if the model supports vision (image/video) inputs
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports vision inputs
+        def supports_vision?(model_id)
+          # FIXME: revise
+          false
+        end
+
+        # Determines if the model supports function calling
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports function calling
+        def supports_functions?(model_id)
+          # FIXME: revise
+          false
+        end
+
+        # Determines if the model supports JSON mode
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports JSON mode
+        def supports_json_mode?(model_id)
+          # FIXME: revise
+          false
+        end
+
+        # Formats the model ID into a human-readable display name
+        # @param model_id [String] the model identifier
+        # @return [String] the formatted display name
+        def format_display_name(model_id)
+          model_id
+            .delete_prefix('models/')
+            .split('-')
+            .map(&:capitalize)
+            .join(' ')
+            .gsub(/(\d+\.\d+)/, ' \1') # Add space before version numbers
+            .gsub(/\s+/, ' ')          # Clean up multiple spaces
+            .strip
+        end
+
+        # Determines if the model supports context caching
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports caching
+        def supports_caching?(model_id)
+          # FIXME: revise
+          true
+        end
+
+        # Determines if the model supports tuning
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports tuning
+        def supports_tuning?(model_id)
+          # FIXME: revise
+          false
+        end
+
+        # Determines if the model supports audio inputs
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports audio inputs
+        def supports_audio?(model_id)
+          # FIXME: revise
+          false
+        end
+
+        # Returns the type of model (chat, embedding, image)
+        # @param model_id [String] the model identifier
+        # @return [String] the model type
+        def model_type(model_id)
+          # FIXME: revise
+          'chat'
+        end
+
+        # Returns the model family identifier
+        # @param model_id [String] the model identifier
+        # @return [String] the model family identifier
+        def model_family(model_id) # rubocop:disable Metrics/CyclomaticComplexity,Metrics/MethodLength
+          'other'
+        end
+
+        # Returns the context length for the model
+        # @param model_id [String] the model identifier
+        # @return [Integer] the context length in tokens
+        def context_length(model_id)
+          context_window_for(model_id)
+        end
+
+        # Default input price for unknown models
+        # @return [Float] the default input price per million tokens
+        def default_input_price
+          0.0
+        end
+
+        # Default output price for unknown models
+        # @return [Float] the default output price per million tokens
+        def default_output_price
+          0.0
+        end
+      end
+    end
+  end
+end