Merge pull request #1006 from alphagov/472-remove-remaining-openai-integration

[CHAT-472] Remove remaining OpenAI integration

Author: davidgisbey

.env.example

@@ -1,14 +1,10 @@
-## variables for working with OpenAI
-OPENAI_ACCESS_TOKEN=
-
 ## Opensearch configuration. Not required for local development using govuk-docker
 OPENSEARCH_URL=
 OPENSEARCH_USERNAME=
 OPENSEARCH_PASSWORD=
 
-## Optional variables to use models hosted on Amazon Bedrock instead of OpenAI
-# ANSWER_STRATEGY=claude_answer_strategy
-# EMBEDDING_PROVIDER=titan
+## Optional variables to use alternate answer strategy
+# ANSWER_STRATEGY=non_llm_answer
 
 ## Optional variables to configure AWS Bedrock interactions
 # BEDROCK_CLAUDE_JAILBREAK_GUARDRAILS_MODEL=

Gemfile

@@ -38,7 +38,6 @@ gem "opensearch-ruby"
 gem "pg"
 gem "prometheus_exporter"
 gem "rack-attack"
-gem "ruby-openai"
 gem "sentry-sidekiq"
 gem "slack-poster"
 gem "sprockets-rails"

Gemfile.lock

@@ -184,7 +184,6 @@ GEM
     ethon (0.18.0)
       ffi (>= 1.15.0)
       logger
-    event_stream_parser (1.0.0)
     execjs (2.10.0)
     factory_bot (6.5.5)
       activesupport (>= 6.1.0)
@@ -197,8 +196,6 @@ GEM
       logger
     faraday-follow_redirects (0.5.0)
       faraday (>= 1, < 3)
-    faraday-multipart (1.1.1)
-      multipart-post (~> 2.0)
     faraday-net_http (3.4.2)
       net-http (~> 0.5)
     faraday-typhoeus (2.0.0)
@@ -400,7 +397,6 @@ GEM
     multi_json (1.19.1)
     multi_xml (0.8.1)
       bigdecimal (>= 3.1, < 5)
-    multipart-post (2.4.1)
     net-http (0.9.1)
       uri (>= 0.11.1)
     net-imap (0.6.3)
@@ -791,10 +787,6 @@ GEM
     rubocop-rspec (3.9.0)
       lint_roller (~> 1.1)
       rubocop (~> 1.81)
-    ruby-openai (8.3.0)
-      event_stream_parser (>= 0.3.0, < 2.0.0)
-      faraday (>= 1)
-      faraday-multipart (>= 1)
     ruby-progressbar (1.13.0)
     rubyzip (2.4.1)
     sass-embedded (1.81.0-aarch64-linux-gnu)
@@ -956,7 +948,6 @@ DEPENDENCIES
   rails (= 8.1.3)
   rspec-rails
   rubocop-govuk
-  ruby-openai
   sentry-sidekiq
   simplecov
   slack-poster
@@ -1027,13 +1018,11 @@ CHECKSUMS
   erb_lint (0.9.0) sha256=dfb5e40ad839e8d1f0d56ca85ec9a7ac4c9cd966ec281138282f35b323ca7c31
   erubi (1.13.1) sha256=a082103b0885dbc5ecf1172fede897f9ebdb745a4b97a5e8dc63953db1ee4ad9
   ethon (0.18.0) sha256=b598afc9f30448cb068b850714b7d6948e941476095d04f90a4ac65b8d6efcb2
-  event_stream_parser (1.0.0) sha256=a2683bab70126286f8184dc88f7968ffc4028f813161fb073ec90d171f7de3c8
   execjs (2.10.0) sha256=6bcb8be8f0052ff9d370b65d1c080f2406656e150452a0abdb185a133048450d
   factory_bot (6.5.5) sha256=ce59295daee1b4704dab8a2fee6816f513d467c6aa3bc587860767d74a66efbe
   factory_bot_rails (6.5.1) sha256=d3cc4851eae4dea8a665ec4a4516895045e710554d2b5ac9e68b94d351bc6d68
   faraday (2.14.1) sha256=a43cceedc1e39d188f4d2cdd360a8aaa6a11da0c407052e426ba8d3fb42ef61c
   faraday-follow_redirects (0.5.0) sha256=5cde93c894b30943a5d2b93c2fe9284216a6b756f7af406a1e55f211d97d10ad
-  faraday-multipart (1.1.1) sha256=77a18ff40149030fd1aef55bb4fc7a67ce46419a8a3fcd010e28c2526e8d8903
   faraday-net_http (3.4.2) sha256=f147758260d3526939bf57ecf911682f94926a3666502e24c69992765875906c
   faraday-typhoeus (2.0.0) sha256=7a459529d782994caf8ce89ca181df3adb89dd471b8ab0b3851936ab566b4dd1
   ffi (1.17.4-aarch64-linux-gnu) sha256=b208f06f91ffd8f5e1193da3cae3d2ccfc27fc36fba577baf698d26d91c080df
@@ -1106,7 +1095,6 @@ CHECKSUMS
   msgpack (1.8.0) sha256=e64ce0212000d016809f5048b48eb3a65ffb169db22238fb4b72472fecb2d732
   multi_json (1.19.1) sha256=7aefeff8f2c854bf739931a238e4aea64592845e0c0395c8a7d2eea7fdd631b7
   multi_xml (0.8.1) sha256=addba0290bac34e9088bfe73dc4878530297a82a7bbd66cb44dcd0a4b86edf5a
-  multipart-post (2.4.1) sha256=9872d03a8e552020ca096adadbf5e3cb1cd1cdd6acd3c161136b8a5737cdb4a8
   net-http (0.9.1) sha256=25ba0b67c63e89df626ed8fac771d0ad24ad151a858af2cc8e6a716ca4336996
   net-imap (0.6.3) sha256=9bab75f876596d09ee7bf911a291da478e0cd6badc54dfb82874855ccc82f2ad
   net-pop (0.1.2) sha256=848b4e982013c15b2f0382792268763b748cce91c9e91e36b0f27ed26420dff3
@@ -1243,7 +1231,6 @@ CHECKSUMS
   rubocop-rails (2.34.3) sha256=10d37989024865ecda8199f311f3faca990143fbac967de943f88aca11eb9ad2
   rubocop-rake (0.7.1) sha256=3797f2b6810c3e9df7376c26d5f44f3475eda59eb1adc38e6f62ecf027cbae4d
   rubocop-rspec (3.9.0) sha256=8fa70a3619408237d789aeecfb9beef40576acc855173e60939d63332fdb55e2
-  ruby-openai (8.3.0) sha256=566dc279c42f4afed68a7a363dce2e594078abfc36b4e043102020b9a387ca69
   ruby-progressbar (1.13.0) sha256=80fc9c47a9b640d6834e0dc7b3c94c9df37f08cb072b7761e4a71e22cff29b33
   rubyzip (2.4.1) sha256=8577c88edc1fde8935eb91064c5cb1aef9ad5494b940cf19c775ee833e075615
   sass-embedded (1.81.0-aarch64-linux-gnu) sha256=9087759cf2f16d0a9263eb97a812dce9882ce79353cfb416f5721835ba2d8706

app.json

@@ -28,9 +28,6 @@
     "AVAILABLE_WITHOUT_SIGNON_AUTHENTICATION": {
       "value": "true"
     },
-    "OPENAI_ACCESS_TOKEN": {
-      "required": true
-    },
     "OPENSEARCH_URL": {
       "required": true
     },

app/models/question.rb

@@ -1,8 +1,6 @@
 class Question < ApplicationRecord
   enum :answer_strategy,
        {
-         open_ai_rag_completion: "open_ai_rag_completion", # legacy strategy - no longer used
-         openai_structured_answer: "openai_structured_answer", # legacy strategy - no longer used
          claude_structured_answer: "claude_structured_answer",
          non_llm_answer: "non_llm_answer", # only used during load testing, but can be present on records created during testing
        },

app/views/static/about.html.erb

@@ -15,16 +15,16 @@
 
     <p class="govuk-body">
       GOV.UK Chat is powered by the same large language model (LLM) used on
-      <%= link_to("OpenAI’s ChatGPT (opens in new tab)",
-                  "https://openai.com/index/chatgpt/",
+      <%= link_to("Anthropic’s Claude (opens in new tab)",
+                  "https://claude.com/product/overview",
                   class: "govuk-link",
                   target: "_blank",
                   rel: "noopener noreferrer") %>.
     </p>
 
     <p class="govuk-body">
-      OpenAI used millions of articles, books and other pieces of text to train
-      the LLM. This is how ChatGPT has the language skills to respond like a human.
+      Anthropic used millions of articles, books and other pieces of text to train
+      the LLM. This is how Claude has the language skills to respond like a human.
     </p>
 
     <p class="govuk-body">
@@ -46,7 +46,7 @@
     <h3 class="govuk-heading-s">Hallucination</h3>
 
     <p class="govuk-body">
-      The biggest limitation of emerging technologies like ChatGPT and GOV.UK
+      The biggest limitation of emerging technologies like Claude and GOV.UK
       Chat is a problem known as hallucination.
     </p>
 

config/application.rb

@@ -66,9 +66,6 @@ class Application < Rails::Application
       api_questions_per_page: 50,
     )
 
-    config.openai_access_token = ENV["OPENAI_ACCESS_TOKEN"]
-    config.openai_request_timeout = 45
-
     config.answer_statuses = Hashie::Mash.new(YAML.load_file("#{__dir__}/answer_statuses.yml"))
     config.question_routing_labels = Hashie::Mash.new(YAML.load_file("#{__dir__}/question_routing_labels.yml"))
     config.search = Hashie::Mash.new(YAML.load_file("#{__dir__}/search.yml", aliases: true))

config/environments/test.rb

@@ -40,8 +40,5 @@
   # Raise error when a before_action's only/except options reference missing actions.
   config.action_controller.raise_on_missing_callback_actions = true
 
-  # Use a hardcoded config value for testing
-  config.openai_access_token = "fake-openai-access-token"
-
   config.conversation_js_progressive_disclosure_delay = 0
 end

config/routes.rb

@@ -10,7 +10,6 @@
     GovukHealthcheck::ActiveRecord,
     GovukHealthcheck::SidekiqRedis,
     GovukHealthcheck::RailsCache,
-    Healthcheck::OpenAI,
     Healthcheck::Opensearch,
     Healthcheck::Bedrock,
   )

docs/guardrails.md

@@ -34,7 +34,7 @@ The output of the LLM is as follows:
 
 We map these to meaningful names using the mappings from a config file, e.g. [here](../config/llm_prompts/answer_guardrails.yml).
 
-The file also contains the prompts we use to run the guardrails. Copy/paste these into the [OpenAI chat playground](https://platform.openai.com/playground/chat?models=gpt-4o) to investigate any issues.
+The file also contains the prompts we use to run the guardrails. Copy/paste these into the [Anthropic workbench](https://platform.claude.com/workbench) to investigate any issues.
 
 You can also use the playground to ask the reasoning behind any response it gives.