Using ruby_llm with little patch json_object. crmne/ruby_llm#11

Author: Eric-Guo

Gemfile

@@ -41,11 +41,12 @@ gem "canonical-rails"
 
 gem "sitemap_generator"
 
-# bundle config local.wechat /Users/guochunzhong/git/oss/wechat/
+# bundle config local.wechat /Users/$(whoami)/git/oss/wechat/
 gem "wechat", git: "https://git.thape.com.cn/Eric-Guo/wechat.git", branch: :main
 gem "rake"
 
-gem "ruby-openai"
+# bundle config local.ruby_llm /Users/$(whoami)/git/oss/ruby_llm/
+gem "ruby_llm", git: "https://git.thape.com.cn/rails/ruby_llm.git", branch: :main
 gem "uri"
 gem "faraday-net_http"
 gem "net-http", "~> 0.4.1"

Gemfile.lock

@@ -12,6 +12,20 @@ GIT
       thor
       zeitwerk (~> 2.4)
 
+GIT
+  remote: https://git.thape.com.cn/rails/ruby_llm.git
+  revision: 32e350578c9e05699fdde67e11e28b01645981ba
+  branch: main
+  specs:
+    ruby_llm (1.1.2)
+      base64
+      event_stream_parser (~> 1)
+      faraday (~> 2)
+      faraday-multipart (~> 1)
+      faraday-net_http (~> 3)
+      faraday-retry (~> 2)
+      zeitwerk (~> 2)
+
 GEM
   remote: https://rubygems.org/
   specs:
@@ -176,6 +190,8 @@ GEM
       multipart-post (~> 2.0)
     faraday-net_http (3.3.0)
       net-http
+    faraday-retry (2.3.1)
+      faraday (~> 2.0)
     ffi (1.16.3)
     ffi-compiler (1.3.2)
       ffi (>= 1.15.5)
@@ -388,10 +404,6 @@ GEM
       sorbet-runtime (>= 0.5.10782)
     ruby-lsp-rails (0.4.0)
       ruby-lsp (>= 0.23.0, < 0.24.0)
-    ruby-openai (8.1.0)
-      event_stream_parser (>= 0.3.0, < 2.0.0)
-      faraday (>= 1)
-      faraday-multipart (>= 1)
     ruby-progressbar (1.13.0)
     ruby-vips (2.2.3)
       ffi (~> 1.12)
@@ -509,7 +521,7 @@ DEPENDENCIES
   rexml
   ruby-lsp
   ruby-lsp-rails
-  ruby-openai
+  ruby_llm!
   securerandom (~> 0.3.2)
   selenium-webdriver (= 4.26.0)
   shakapacker (~> 8.2)

app/models/guest_message.rb

@@ -59,15 +59,9 @@ def ai_spam_score
 }
 EOS_PROMPT
 
-    response = OpenAI::Client.new.chat(
-      parameters: {
-        model: "deepseek-chat",
-        response_format: {type: "json_object"},
-        messages: [{role: "system", content: system_prompt},
-          {role: "user", content: "#{name} from company #{company} (#{contact_details}) leaving message:\n\n#{message}"}],
-        temperature: 0.7
-      }
-    )
-    response&.dig("choices", 0, "message", "content")
+    chat = RubyLLM.chat
+    chat.with_instructions system_prompt
+    response = chat.ask "#{name} from company #{company} (#{contact_details}) leaving message:\n\n#{message}"
+    response.content
   end
 end

config/initializers/openai.rb

@@ -0,0 +1,37 @@
+require "ruby_llm"
+
+RubyLLM.configure do |config|
+  # --- Provider API Keys ---
+  # Provide keys ONLY for the providers you intend to use.
+  # Using environment variables (ENV.fetch) is highly recommended.
+  config.openai_api_key = ENV.fetch("OPENAI_API_KEY", nil)
+  config.anthropic_api_key = ENV.fetch("ANTHROPIC_API_KEY", nil)
+  config.gemini_api_key = ENV.fetch("GEMINI_API_KEY", nil)
+  config.deepseek_api_key = ENV.fetch("DEEPSEEK_API_KEY", Rails.application.credentials.deepseek_access_token)
+
+  # --- AWS Bedrock Credentials ---
+  # Uses standard AWS credential chain (environment, shared config, IAM role)
+  # if these specific keys aren't set. Region is required if using Bedrock.
+  config.bedrock_api_key = ENV.fetch("AWS_ACCESS_KEY_ID", nil)
+  config.bedrock_secret_key = ENV.fetch("AWS_SECRET_ACCESS_KEY", nil)
+  config.bedrock_region = ENV.fetch("AWS_REGION", nil) # e.g., 'us-west-2'
+  config.bedrock_session_token = ENV.fetch("AWS_SESSION_TOKEN", nil) # For temporary credentials
+
+  # --- Custom OpenAI Endpoint --- New in v1.2.0
+  # Use this for Azure OpenAI, proxies, or self-hosted models via OpenAI-compatible APIs.
+  # See the "Working with Models" guide for details.
+  config.openai_api_base = ENV.fetch("OPENAI_API_BASE", nil) # e.g., "https://your-azure.openai.azure.com"
+
+  # --- Default Models ---
+  # Used by RubyLLM.chat, RubyLLM.embed, RubyLLM.paint if no model is specified.
+  config.default_model = "deepseek-chat"               # Default: 'gpt-4.1-nano'
+  config.default_embedding_model = "text-embedding-3-small"  # Default: 'text-embedding-3-small'
+  config.default_image_model = "dall-e-3"            # Default: 'dall-e-3'
+
+  # --- Connection Settings ---
+  config.request_timeout = 60  # Request timeout in seconds (default: 120)
+  config.max_retries = 2        # Max retries on transient network errors (default: 3)
+  config.retry_interval = 0.1 # Initial delay in seconds (default: 0.1)
+  config.retry_backoff_factor = 2 # Multiplier for subsequent retries (default: 2)
+  config.retry_interval_randomness = 0.5 # Jitter factor (default: 0.5)
+end