diff --git a/.gitignore b/.gitignore
index 3b519490..e0744210 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,3 +2,5 @@ node_modules
 /gems
 /auto_generated
 .env
+evals/log
+evals/cases
diff --git a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-edit.js b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-edit.js
index 7ba40586..65b3f7f3 100644
--- a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-edit.js
+++ b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-edit.js
@@ -14,5 +14,7 @@ export default class DiscourseAiToolsEditRoute extends DiscourseRoute {
 
     controller.set("allTools", toolsModel);
     controller.set("presets", toolsModel.resultSetMeta.presets);
+    controller.set("llms", toolsModel.resultSetMeta.llms);
+    controller.set("settings", toolsModel.resultSetMeta.settings);
   }
 }
diff --git a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-new.js b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-new.js
index 5c05907f..1e645097 100644
--- a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-new.js
+++ b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-tools-new.js
@@ -11,5 +11,7 @@ export default class DiscourseAiToolsNewRoute extends DiscourseRoute {
 
     controller.set("allTools", toolsModel);
     controller.set("presets", toolsModel.resultSetMeta.presets);
+    controller.set("llms", toolsModel.resultSetMeta.llms);
+    controller.set("settings", toolsModel.resultSetMeta.settings);
   }
 }
diff --git a/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/edit.hbs b/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/edit.hbs
index 7fcc45c5..444f6966 100644
--- a/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/edit.hbs
+++ b/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/edit.hbs
@@ -3,5 +3,7 @@
     @tools={{this.allTools}}
     @model={{this.model}}
     @presets={{this.presets}}
+    @llms={{this.llms}}
+    @settings={{this.settings}}
   />
 </section>
\ No newline at end of file
diff --git a/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/new.hbs b/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/new.hbs
index 7fcc45c5..444f6966 100644
--- a/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/new.hbs
+++ b/admin/assets/javascripts/discourse/templates/admin-plugins/show/discourse-ai-tools/new.hbs
@@ -3,5 +3,7 @@
     @tools={{this.allTools}}
     @model={{this.model}}
     @presets={{this.presets}}
+    @llms={{this.llms}}
+    @settings={{this.settings}}
   />
 </section>
\ No newline at end of file
diff --git a/app/controllers/discourse_ai/admin/ai_personas_controller.rb b/app/controllers/discourse_ai/admin/ai_personas_controller.rb
index b0317f02..540ac824 100644
--- a/app/controllers/discourse_ai/admin/ai_personas_controller.rb
+++ b/app/controllers/discourse_ai/admin/ai_personas_controller.rb
@@ -32,10 +32,19 @@ module DiscourseAi
             }
           end
         llms =
-          DiscourseAi::Configuration::LlmEnumerator
-            .values(allowed_seeded_llms: SiteSetting.ai_bot_allowed_seeded_models)
-            .map { |hash| { id: hash[:value], name: hash[:name] } }
-        render json: { ai_personas: ai_personas, meta: { tools: tools, llms: llms } }
+          DiscourseAi::Configuration::LlmEnumerator.values_for_serialization(
+            allowed_seeded_llm_ids: SiteSetting.ai_bot_allowed_seeded_models_map,
+          )
+        render json: {
+                 ai_personas: ai_personas,
+                 meta: {
+                   tools: tools,
+                   llms: llms,
+                   settings: {
+                     rag_pdf_images_enabled: SiteSetting.ai_rag_pdf_images_enabled,
+                   },
+                 },
+               }
       end
 
       def new
@@ -187,7 +196,7 @@ module DiscourseAi
             :priority,
             :top_p,
             :temperature,
-            :default_llm,
+            :default_llm_id,
             :user_id,
             :max_context_posts,
             :vision_enabled,
@@ -195,7 +204,8 @@ module DiscourseAi
             :rag_chunk_tokens,
             :rag_chunk_overlap_tokens,
             :rag_conversation_chunks,
-            :question_consolidator_llm,
+            :rag_llm_model_id,
+            :question_consolidator_llm_id,
             :allow_chat_channel_mentions,
             :allow_chat_direct_messages,
             :allow_topic_mentions,
diff --git a/app/controllers/discourse_ai/admin/ai_tools_controller.rb b/app/controllers/discourse_ai/admin/ai_tools_controller.rb
index fc7b2984..caf7fae6 100644
--- a/app/controllers/discourse_ai/admin/ai_tools_controller.rb
+++ b/app/controllers/discourse_ai/admin/ai_tools_controller.rb
@@ -90,6 +90,7 @@ module DiscourseAi
             :summary,
             :rag_chunk_tokens,
             :rag_chunk_overlap_tokens,
+            :rag_llm_model_id,
             rag_uploads: [:id],
             parameters: [:name, :type, :description, :required, enum: []],
           )
diff --git a/app/controllers/discourse_ai/admin/rag_document_fragments_controller.rb b/app/controllers/discourse_ai/admin/rag_document_fragments_controller.rb
index 732a28fb..d27f4a82 100644
--- a/app/controllers/discourse_ai/admin/rag_document_fragments_controller.rb
+++ b/app/controllers/discourse_ai/admin/rag_document_fragments_controller.rb
@@ -49,6 +49,7 @@ module DiscourseAi
       def validate_extension!(filename)
         extension = File.extname(filename)[1..-1] || ""
         authorized_extensions = %w[txt md]
+        authorized_extensions.concat(%w[pdf png jpg jpeg]) if SiteSetting.ai_rag_pdf_images_enabled
         if !authorized_extensions.include?(extension)
           raise Discourse::InvalidParameters.new(
                   I18n.t(
diff --git a/app/jobs/regular/digest_rag_upload.rb b/app/jobs/regular/digest_rag_upload.rb
index bfc2ac4b..179660d1 100644
--- a/app/jobs/regular/digest_rag_upload.rb
+++ b/app/jobs/regular/digest_rag_upload.rb
@@ -28,7 +28,7 @@ module ::Jobs
 
       # Check if this is the first time we process this upload.
       if fragment_ids.empty?
-        document = get_uploaded_file(upload)
+        document = get_uploaded_file(upload: upload, target: target)
         return if document.nil?
 
         RagDocumentFragment.publish_status(upload, { total: 0, indexed: 0, left: 0 })
@@ -163,7 +163,38 @@ module ::Jobs
       [buffer, split_char]
     end
 
-    def get_uploaded_file(upload)
+    def get_uploaded_file(upload:, target:)
+      if %w[pdf png jpg jpeg].include?(upload.extension) && !SiteSetting.ai_rag_pdf_images_enabled
+        raise Discourse::InvalidAccess.new(
+                "The setting ai_rag_pdf_images_enabled is false, can not index images and pdfs.",
+              )
+      end
+      if upload.extension == "pdf"
+        pages =
+          DiscourseAi::Utils::PdfToImages.new(
+            upload: upload,
+            user: Discourse.system_user,
+          ).uploaded_pages
+
+        return(
+          DiscourseAi::Utils::ImageToText.as_fake_file(
+            uploads: pages,
+            llm_model: target.rag_llm_model,
+            user: Discourse.system_user,
+          )
+        )
+      end
+
+      if %w[png jpg jpeg].include?(upload.extension)
+        return(
+          DiscourseAi::Utils::ImageToText.as_fake_file(
+            uploads: [upload],
+            llm_model: target.rag_llm_model,
+            user: Discourse.system_user,
+          )
+        )
+      end
+
       store = Discourse.store
       @file ||=
         if store.external?
diff --git a/app/models/ai_persona.rb b/app/models/ai_persona.rb
index 2c573977..867d44ce 100644
--- a/app/models/ai_persona.rb
+++ b/app/models/ai_persona.rb
@@ -1,8 +1,8 @@
 # frozen_string_literal: true
 
 class AiPersona < ActiveRecord::Base
-  # TODO remove this line 01-1-2025
-  self.ignored_columns = %i[commands allow_chat mentionable]
+  # TODO remove this line 01-10-2025
+  self.ignored_columns = %i[default_llm question_consolidator_llm]
 
   # places a hard limit, so per site we cache a maximum of 500 classes
   MAX_PERSONAS_PER_SITE = 500
@@ -12,7 +12,7 @@ class AiPersona < ActiveRecord::Base
   validates :system_prompt, presence: true, length: { maximum: 10_000_000 }
   validate :system_persona_unchangeable, on: :update, if: :system
   validate :chat_preconditions
-  validate :allowed_seeded_model, if: :default_llm
+  validate :allowed_seeded_model, if: :default_llm_id
   validates :max_context_posts, numericality: { greater_than: 0 }, allow_nil: true
   # leaves some room for growth but sets a maximum to avoid memory issues
   # we may want to revisit this in the future
@@ -30,6 +30,10 @@ class AiPersona < ActiveRecord::Base
   belongs_to :created_by, class_name: "User"
   belongs_to :user
 
+  belongs_to :default_llm, class_name: "LlmModel"
+  belongs_to :question_consolidator_llm, class_name: "LlmModel"
+  belongs_to :rag_llm_model, class_name: "LlmModel"
+
   has_many :upload_references, as: :target, dependent: :destroy
   has_many :uploads, through: :upload_references
 
@@ -62,7 +66,7 @@ class AiPersona < ActiveRecord::Base
             user_id: persona.user_id,
             username: persona.user.username_lower,
             allowed_group_ids: persona.allowed_group_ids,
-            default_llm: persona.default_llm,
+            default_llm_id: persona.default_llm_id,
             force_default_llm: persona.force_default_llm,
             allow_chat_channel_mentions: persona.allow_chat_channel_mentions,
             allow_chat_direct_messages: persona.allow_chat_direct_messages,
@@ -157,12 +161,12 @@ class AiPersona < ActiveRecord::Base
       user_id
       system
       mentionable
-      default_llm
+      default_llm_id
       max_context_posts
       vision_enabled
       vision_max_pixels
       rag_conversation_chunks
-      question_consolidator_llm
+      question_consolidator_llm_id
       allow_chat_channel_mentions
       allow_chat_direct_messages
       allow_topic_mentions
@@ -302,7 +306,7 @@ class AiPersona < ActiveRecord::Base
     if (
          allow_chat_channel_mentions || allow_chat_direct_messages || allow_topic_mentions ||
            force_default_llm
-       ) && !default_llm
+       ) && !default_llm_id
       errors.add(:default_llm, I18n.t("discourse_ai.ai_bot.personas.default_llm_required"))
     end
   end
@@ -332,13 +336,12 @@ class AiPersona < ActiveRecord::Base
   end
 
   def allowed_seeded_model
-    return if default_llm.blank?
+    return if default_llm_id.blank?
 
-    llm = LlmModel.find_by(id: default_llm.split(":").last.to_i)
-    return if llm.nil?
-    return if !llm.seeded?
+    return if default_llm.nil?
+    return if !default_llm.seeded?
 
-    return if SiteSetting.ai_bot_allowed_seeded_models.include?(llm.id.to_s)
+    return if SiteSetting.ai_bot_allowed_seeded_models_map.include?(default_llm.id.to_s)
 
     errors.add(:default_llm, I18n.t("discourse_ai.llm.configuration.invalid_seeded_model"))
   end
@@ -348,36 +351,37 @@ end
 #
 # Table name: ai_personas
 #
-#  id                          :bigint           not null, primary key
-#  name                        :string(100)      not null
-#  description                 :string(2000)     not null
-#  system_prompt               :string(10000000) not null
-#  allowed_group_ids           :integer          default([]), not null, is an Array
-#  created_by_id               :integer
-#  enabled                     :boolean          default(TRUE), not null
-#  created_at                  :datetime         not null
-#  updated_at                  :datetime         not null
-#  system                      :boolean          default(FALSE), not null
-#  priority                    :boolean          default(FALSE), not null
-#  temperature                 :float
-#  top_p                       :float
-#  user_id                     :integer
-#  default_llm                 :text
-#  max_context_posts           :integer
-#  vision_enabled              :boolean          default(FALSE), not null
-#  vision_max_pixels           :integer          default(1048576), not null
-#  rag_chunk_tokens            :integer          default(374), not null
-#  rag_chunk_overlap_tokens    :integer          default(10), not null
-#  rag_conversation_chunks     :integer          default(10), not null
-#  question_consolidator_llm   :text
-#  tool_details                :boolean          default(TRUE), not null
-#  tools                       :json             not null
-#  forced_tool_count           :integer          default(-1), not null
-#  allow_chat_channel_mentions :boolean          default(FALSE), not null
-#  allow_chat_direct_messages  :boolean          default(FALSE), not null
-#  allow_topic_mentions        :boolean          default(FALSE), not null
-#  allow_personal_messages     :boolean          default(TRUE), not null
-#  force_default_llm           :boolean          default(FALSE), not null
+#  id                           :bigint           not null, primary key
+#  name                         :string(100)      not null
+#  description                  :string(2000)     not null
+#  system_prompt                :string(10000000) not null
+#  allowed_group_ids            :integer          default([]), not null, is an Array
+#  created_by_id                :integer
+#  enabled                      :boolean          default(TRUE), not null
+#  created_at                   :datetime         not null
+#  updated_at                   :datetime         not null
+#  system                       :boolean          default(FALSE), not null
+#  priority                     :boolean          default(FALSE), not null
+#  temperature                  :float
+#  top_p                        :float
+#  user_id                      :integer
+#  max_context_posts            :integer
+#  vision_enabled               :boolean          default(FALSE), not null
+#  vision_max_pixels            :integer          default(1048576), not null
+#  rag_chunk_tokens             :integer          default(374), not null
+#  rag_chunk_overlap_tokens     :integer          default(10), not null
+#  rag_conversation_chunks      :integer          default(10), not null
+#  tool_details                 :boolean          default(TRUE), not null
+#  tools                        :json             not null
+#  forced_tool_count            :integer          default(-1), not null
+#  allow_chat_channel_mentions  :boolean          default(FALSE), not null
+#  allow_chat_direct_messages   :boolean          default(FALSE), not null
+#  allow_topic_mentions         :boolean          default(FALSE), not null
+#  allow_personal_messages      :boolean          default(TRUE), not null
+#  force_default_llm            :boolean          default(FALSE), not null
+#  rag_llm_model_id             :bigint
+#  default_llm_id               :bigint
+#  question_consolidator_llm_id :bigint
 #
 # Indexes
 #
diff --git a/app/models/ai_tool.rb b/app/models/ai_tool.rb
index 97b2a983..ba3b4098 100644
--- a/app/models/ai_tool.rb
+++ b/app/models/ai_tool.rb
@@ -8,6 +8,7 @@ class AiTool < ActiveRecord::Base
   validates :script, presence: true, length: { maximum: 100_000 }
   validates :created_by_id, presence: true
   belongs_to :created_by, class_name: "User"
+  belongs_to :rag_llm_model, class_name: "LlmModel"
   has_many :rag_document_fragments, dependent: :destroy, as: :target
   has_many :upload_references, as: :target, dependent: :destroy
   has_many :uploads, through: :upload_references
@@ -371,4 +372,4 @@ end
 #  rag_chunk_tokens         :integer          default(374), not null
 #  rag_chunk_overlap_tokens :integer          default(10), not null
 #  tool_name                :string(100)      default(""), not null
-#
+#  rag_llm_model_id         :bigint
diff --git a/app/models/llm_model.rb b/app/models/llm_model.rb
index 228c9335..a5f5e510 100644
--- a/app/models/llm_model.rb
+++ b/app/models/llm_model.rb
@@ -70,7 +70,7 @@ class LlmModel < ActiveRecord::Base
   end
 
   def to_llm
-    DiscourseAi::Completions::Llm.proxy(identifier)
+    DiscourseAi::Completions::Llm.proxy(self)
   end
 
   def identifier
diff --git a/app/serializers/ai_custom_tool_list_serializer.rb b/app/serializers/ai_custom_tool_list_serializer.rb
index f0e3f1cc..9f4ad25d 100644
--- a/app/serializers/ai_custom_tool_list_serializer.rb
+++ b/app/serializers/ai_custom_tool_list_serializer.rb
@@ -6,7 +6,13 @@ class AiCustomToolListSerializer < ApplicationSerializer
   has_many :ai_tools, serializer: AiCustomToolSerializer, embed: :objects
 
   def meta
-    { presets: AiTool.presets }
+    {
+      presets: AiTool.presets,
+      llms: DiscourseAi::Configuration::LlmEnumerator.values_for_serialization,
+      settings: {
+        rag_pdf_images_enabled: SiteSetting.ai_rag_pdf_images_enabled,
+      },
+    }
   end
 
   def ai_tools
diff --git a/app/serializers/ai_custom_tool_serializer.rb b/app/serializers/ai_custom_tool_serializer.rb
index 1ac80430..2af1a89f 100644
--- a/app/serializers/ai_custom_tool_serializer.rb
+++ b/app/serializers/ai_custom_tool_serializer.rb
@@ -10,6 +10,7 @@ class AiCustomToolSerializer < ApplicationSerializer
              :script,
              :rag_chunk_tokens,
              :rag_chunk_overlap_tokens,
+             :rag_llm_model_id,
              :created_by_id,
              :created_at,
              :updated_at
diff --git a/app/serializers/localized_ai_persona_serializer.rb b/app/serializers/localized_ai_persona_serializer.rb
index 81f6fa85..dde41dfb 100644
--- a/app/serializers/localized_ai_persona_serializer.rb
+++ b/app/serializers/localized_ai_persona_serializer.rb
@@ -14,7 +14,7 @@ class LocalizedAiPersonaSerializer < ApplicationSerializer
              :allowed_group_ids,
              :temperature,
              :top_p,
-             :default_llm,
+             :default_llm_id,
              :user_id,
              :max_context_posts,
              :vision_enabled,
@@ -22,7 +22,8 @@ class LocalizedAiPersonaSerializer < ApplicationSerializer
              :rag_chunk_tokens,
              :rag_chunk_overlap_tokens,
              :rag_conversation_chunks,
-             :question_consolidator_llm,
+             :rag_llm_model_id,
+             :question_consolidator_llm_id,
              :tool_details,
              :forced_tool_count,
              :allow_chat_channel_mentions,
diff --git a/assets/javascripts/discourse/admin/models/ai-persona.js b/assets/javascripts/discourse/admin/models/ai-persona.js
index 2c9c7383..18c97a7d 100644
--- a/assets/javascripts/discourse/admin/models/ai-persona.js
+++ b/assets/javascripts/discourse/admin/models/ai-persona.js
@@ -15,7 +15,7 @@ const CREATE_ATTRIBUTES = [
   "top_p",
   "temperature",
   "user_id",
-  "default_llm",
+  "default_llm_id",
   "force_default_llm",
   "user",
   "max_context_posts",
@@ -25,7 +25,8 @@ const CREATE_ATTRIBUTES = [
   "rag_chunk_tokens",
   "rag_chunk_overlap_tokens",
   "rag_conversation_chunks",
-  "question_consolidator_llm",
+  "rag_llm_model_id",
+  "question_consolidator_llm_id",
   "allow_chat",
   "tool_details",
   "forced_tool_count",
@@ -43,7 +44,7 @@ const SYSTEM_ATTRIBUTES = [
   "priority",
   "tools",
   "user_id",
-  "default_llm",
+  "default_llm_id",
   "force_default_llm",
   "user",
   "max_context_posts",
@@ -53,7 +54,8 @@ const SYSTEM_ATTRIBUTES = [
   "rag_chunk_tokens",
   "rag_chunk_overlap_tokens",
   "rag_conversation_chunks",
-  "question_consolidator_llm",
+  "rag_llm_model_id",
+  "question_consolidator_llm_id",
   "tool_details",
   "allow_personal_messages",
   "allow_topic_mentions",
diff --git a/assets/javascripts/discourse/admin/models/ai-tool.js b/assets/javascripts/discourse/admin/models/ai-tool.js
index 7188a68c..4b9cdf7c 100644
--- a/assets/javascripts/discourse/admin/models/ai-tool.js
+++ b/assets/javascripts/discourse/admin/models/ai-tool.js
@@ -12,6 +12,7 @@ const CREATE_ATTRIBUTES = [
   "rag_uploads",
   "rag_chunk_tokens",
   "rag_chunk_overlap_tokens",
+  "rag_llm_model_id",
   "enabled",
 ];
 
diff --git a/assets/javascripts/discourse/components/ai-persona-editor.gjs b/assets/javascripts/discourse/components/ai-persona-editor.gjs
index 121ad852..2f4c1bd8 100644
--- a/assets/javascripts/discourse/components/ai-persona-editor.gjs
+++ b/assets/javascripts/discourse/components/ai-persona-editor.gjs
@@ -167,27 +167,27 @@ export default class PersonaEditor extends Component {
   }
 
   get mappedQuestionConsolidatorLlm() {
-    return this.editingModel?.question_consolidator_llm || "blank";
+    return this.editingModel?.question_consolidator_llm_id ?? "blank";
   }
 
   set mappedQuestionConsolidatorLlm(value) {
     if (value === "blank") {
-      this.editingModel.question_consolidator_llm = null;
+      this.editingModel.question_consolidator_llm_id = null;
     } else {
-      this.editingModel.question_consolidator_llm = value;
+      this.editingModel.question_consolidator_llm_id = value;
     }
   }
 
   get mappedDefaultLlm() {
-    return this.editingModel?.default_llm || "blank";
+    return this.editingModel?.default_llm_id ?? "blank";
   }
 
   set mappedDefaultLlm(value) {
     if (value === "blank") {
-      this.editingModel.default_llm = null;
+      this.editingModel.default_llm_id = null;
       this.hasDefaultLlm = false;
     } else {
-      this.editingModel.default_llm = value;
+      this.editingModel.default_llm_id = value;
       this.hasDefaultLlm = true;
     }
   }
@@ -596,9 +596,14 @@ export default class PersonaEditor extends Component {
             @target={{this.editingModel}}
             @updateUploads={{this.updateUploads}}
             @onRemove={{this.removeUpload}}
+            @allowPdfsAndImages={{@personas.resultSetMeta.settings.rag_pdf_images_enabled}}
           />
         </div>
-        <RagOptions @model={{this.editingModel}}>
+        <RagOptions
+          @model={{this.editingModel}}
+          @llms={{@personas.resultSetMeta.llms}}
+          @allowPdfsAndImages={{@personas.resultSetMeta.settings.rag_pdf_images_enabled}}
+        >
           <div class="control-group">
             <label>{{i18n
                 "discourse_ai.ai_persona.rag_conversation_chunks"
diff --git a/assets/javascripts/discourse/components/ai-tool-editor.gjs b/assets/javascripts/discourse/components/ai-tool-editor.gjs
index 154ce390..94d4db82 100644
--- a/assets/javascripts/discourse/components/ai-tool-editor.gjs
+++ b/assets/javascripts/discourse/components/ai-tool-editor.gjs
@@ -90,7 +90,8 @@ export default class AiToolEditor extends Component {
         "summary",
         "rag_uploads",
         "rag_chunk_tokens",
-        "rag_chunk_overlap_tokens"
+        "rag_chunk_overlap_tokens",
+        "rag_llm_model_id"
       );
 
       await this.args.model.save(data);
@@ -244,9 +245,14 @@ export default class AiToolEditor extends Component {
               @target={{this.editingModel}}
               @updateUploads={{this.updateUploads}}
               @onRemove={{this.removeUpload}}
+              @allowPdfsAndImages={{@settings.rag_pdf_images_enabled}}
             />
           </div>
-          <RagOptions @model={{this.editingModel}} />
+          <RagOptions
+            @model={{this.editingModel}}
+            @llms={{@llms}}
+            @allowPdfsAndImages={{@settings.rag_pdf_images_enabled}}
+          />
         {{/if}}
 
         <div class="control-group ai-tool-editor__action_panel">
diff --git a/assets/javascripts/discourse/components/rag-options.gjs b/assets/javascripts/discourse/components/rag-options.gjs
index 894c1895..2f38346b 100644
--- a/assets/javascripts/discourse/components/rag-options.gjs
+++ b/assets/javascripts/discourse/components/rag-options.gjs
@@ -5,6 +5,7 @@ import { on } from "@ember/modifier";
 import { action } from "@ember/object";
 import DTooltip from "discourse/components/d-tooltip";
 import { i18n } from "discourse-i18n";
+import AiLlmSelector from "./ai-llm-selector";
 
 export default class RagOptions extends Component {
   @tracked showIndexingOptions = false;
@@ -22,6 +23,22 @@ export default class RagOptions extends Component {
       : i18n("discourse_ai.rag.options.show_indexing_options");
   }
 
+  get visionLlms() {
+    return this.args.llms.filter((llm) => llm.vision_enabled);
+  }
+
+  get visionLlmId() {
+    return this.args.model.rag_llm_model_id ?? "blank";
+  }
+
+  set visionLlmId(value) {
+    if (value === "blank") {
+      this.args.model.rag_llm_model_id = null;
+    } else {
+      this.args.model.rag_llm_model_id = value;
+    }
+  }
+
   <template>
     {{#if @model.rag_uploads}}
       <a
@@ -64,6 +81,20 @@ export default class RagOptions extends Component {
           }}
         />
       </div>
+      {{#if @allowPdfsAndImages}}
+        <div class="control-group">
+          <label>{{i18n "discourse_ai.rag.options.rag_llm_model"}}</label>
+          <AiLlmSelector
+            class="ai-persona-editor__llms"
+            @value={{this.visionLlmId}}
+            @llms={{this.visionLlms}}
+          />
+          <DTooltip
+            @icon="circle-question"
+            @content={{i18n "discourse_ai.rag.options.rag_llm_model_help"}}
+          />
+        </div>
+      {{/if}}
       {{yield}}
     {{/if}}
   </template>
diff --git a/assets/javascripts/discourse/components/rag-uploader.gjs b/assets/javascripts/discourse/components/rag-uploader.gjs
index 115e9293..510f003b 100644
--- a/assets/javascripts/discourse/components/rag-uploader.gjs
+++ b/assets/javascripts/discourse/components/rag-uploader.gjs
@@ -77,6 +77,14 @@ export default class RagUploader extends Component {
     this.updateUploads(this.ragUploads);
   }
 
+  get acceptedFileTypes() {
+    if (this.args?.allowPdfsAndImages) {
+      return ".txt,.md,.pdf,.png,.jpg,.jpeg";
+    } else {
+      return ".txt,.md";
+    }
+  }
+
   @action
   submitFiles() {
     this.uppyUpload.openPicker();
@@ -119,7 +127,11 @@ export default class RagUploader extends Component {
   <template>
     <div class="rag-uploader">
       <h3>{{i18n "discourse_ai.rag.uploads.title"}}</h3>
-      <p>{{i18n "discourse_ai.rag.uploads.description"}}</p>
+      {{#if @allowPdfsAndImages}}
+        <p>{{i18n "discourse_ai.rag.uploads.description_with_pdfs"}}</p>
+      {{else}}
+        <p>{{i18n "discourse_ai.rag.uploads.description"}}</p>
+      {{/if}}
 
       {{#if this.ragUploads}}
         <div class="rag-uploader__search-input-container">
@@ -187,7 +199,7 @@ export default class RagUploader extends Component {
         disabled={{this.uploading}}
         type="file"
         multiple="multiple"
-        accept=".txt,.md"
+        accept={{this.acceptedFileTypes}}
       />
       <DButton
         @label="discourse_ai.rag.uploads.button"
diff --git a/config/locales/client.en.yml b/config/locales/client.en.yml
index 1bdb0cca..e738b4bb 100644
--- a/config/locales/client.en.yml
+++ b/config/locales/client.en.yml
@@ -274,11 +274,14 @@ en:
           rag_chunk_tokens_help: "The number of tokens to use for each chunk in the RAG model. Increase to increase the amount of context the AI can use. (changing will re-index all uploads)"
           rag_chunk_overlap_tokens: "Upload chunk overlap tokens"
           rag_chunk_overlap_tokens_help: "The number of tokens to overlap between chunks in the RAG model. (changing will re-index all uploads)"
+          rag_llm_model: "Indexing Language Model"
+          rag_llm_model_help: "The language model used for OCR during indexing of PDFs and images"
           show_indexing_options: "Show upload options"
           hide_indexing_options: "Hide upload options"
         uploads:
           title: "Uploads"
-          description: "Uploaded files should be formatted as plaintext (.txt) or markdown (.md)."
+          description: "Plaintext (.txt) or markdown (.md)"
+          description_with_pdfs: "Plaintext (.txt), markdown (.md), PDF (.pdf) or image (.png, .jpeg)"
           button: "Add files"
           filter: "Filter uploads"
           indexed: "Indexed"
diff --git a/config/settings.yml b/config/settings.yml
index 1d95e7a4..850d63ec 100644
--- a/config/settings.yml
+++ b/config/settings.yml
@@ -347,12 +347,14 @@ discourse_ai:
   ai_spam_detection_enabled:
     default: false
     hidden: true
-
   ai_spam_detection_user_id:
     default: ""
     hidden: true
-
   ai_spam_detection_model_allowed_seeded_models:
     default: ""
     hidden: true
     type: list
+
+  ai_rag_pdf_images_enabled:
+    default: false
+    hidden: true
diff --git a/db/migrate/20250210024600_add_rag_llm_model.rb b/db/migrate/20250210024600_add_rag_llm_model.rb
new file mode 100644
index 00000000..bcf8ad79
--- /dev/null
+++ b/db/migrate/20250210024600_add_rag_llm_model.rb
@@ -0,0 +1,7 @@
+# frozen_string_literal: true
+class AddRagLlmModel < ActiveRecord::Migration[7.2]
+  def change
+    add_column :ai_personas, :rag_llm_model_id, :bigint
+    add_column :ai_tools, :rag_llm_model_id, :bigint
+  end
+end
diff --git a/db/migrate/20250210032345_migrate_persona_to_llm_model_id.rb b/db/migrate/20250210032345_migrate_persona_to_llm_model_id.rb
new file mode 100644
index 00000000..4c601703
--- /dev/null
+++ b/db/migrate/20250210032345_migrate_persona_to_llm_model_id.rb
@@ -0,0 +1,19 @@
+# frozen_string_literal: true
+class MigratePersonaToLlmModelId < ActiveRecord::Migration[7.2]
+  def up
+    add_column :ai_personas, :default_llm_id, :bigint
+    add_column :ai_personas, :question_consolidator_llm_id, :bigint
+    # personas are seeded, we do not mark stuff as readonline
+
+    execute <<~SQL
+      UPDATE ai_personas
+        set
+          default_llm_id = (select id from llm_models where ('custom:' || id) = default_llm),
+          question_consolidator_llm_id = (select id from llm_models where ('custom:' || id) = question_consolidator_llm)
+      SQL
+  end
+
+  def down
+    raise ActiveRecord::IrreversibleMigration
+  end
+end
diff --git a/db/post_migrate/20250210032351_post_migrate_persona_to_llm_model_id.rb b/db/post_migrate/20250210032351_post_migrate_persona_to_llm_model_id.rb
new file mode 100644
index 00000000..0d161b56
--- /dev/null
+++ b/db/post_migrate/20250210032351_post_migrate_persona_to_llm_model_id.rb
@@ -0,0 +1,11 @@
+# frozen_string_literal: true
+class PostMigratePersonaToLlmModelId < ActiveRecord::Migration[7.2]
+  def up
+    remove_column :ai_personas, :default_llm
+    remove_column :ai_personas, :question_consolidator_llm
+  end
+
+  def down
+    raise ActiveRecord::IrreversibleMigration
+  end
+end
diff --git a/evals/lib/boot.rb b/evals/lib/boot.rb
new file mode 100644
index 00000000..15144e90
--- /dev/null
+++ b/evals/lib/boot.rb
@@ -0,0 +1,36 @@
+# frozen_string_literal: true
+
+# got to ensure evals are here
+# rubocop:disable Discourse/Plugins/NamespaceConstants
+EVAL_PATH = File.join(__dir__, "../cases")
+# rubocop:enable Discourse/Plugins/NamespaceConstants
+#
+if !Dir.exist?(EVAL_PATH)
+  puts "Evals are missing, cloning from discourse/discourse-ai-evals"
+
+  success =
+    system("git clone git@github.com:discourse/discourse-ai-evals.git '#{EVAL_PATH}' 2>/dev/null")
+
+  # Fall back to HTTPS if SSH fails
+  if !success
+    puts "SSH clone failed, falling back to HTTPS..."
+    success = system("git clone https://github.com/discourse/discourse-ai-evals.git '#{EVAL_PATH}'")
+  end
+
+  if success
+    puts "Successfully cloned evals repository"
+  else
+    abort "Failed to clone evals repository"
+  end
+end
+
+discourse_path = File.expand_path(File.join(__dir__, "../../../.."))
+# rubocop:disable Discourse/NoChdir
+Dir.chdir(discourse_path)
+# rubocop:enable Discourse/NoChdir
+
+require "/home/sam/Source/discourse/config/environment"
+
+ENV["DISCOURSE_AI_NO_DEBUG"] = "1"
+module DiscourseAi::Evals
+end
diff --git a/evals/lib/cli.rb b/evals/lib/cli.rb
new file mode 100644
index 00000000..e28658ea
--- /dev/null
+++ b/evals/lib/cli.rb
@@ -0,0 +1,47 @@
+# frozen_string_literal: true
+require "optparse"
+
+class DiscourseAi::Evals::Cli
+  class Options
+    attr_accessor :eval_name, :model, :list, :list_models
+    def initialize(eval_name: nil, model: nil, list: false, list_models: false)
+      @eval_name = eval_name
+      @model = model
+      @list = list
+      @list_models = list_models
+    end
+  end
+
+  def self.parse_options!
+    options = Options.new
+
+    parser =
+      OptionParser.new do |opts|
+        opts.banner = "Usage: evals/run [options]"
+
+        opts.on("-e", "--eval NAME", "Name of the evaluation to run") do |eval_name|
+          options.eval_name = eval_name
+        end
+
+        opts.on("--list-models", "List models") { |model| options.list_models = true }
+
+        opts.on(
+          "-m",
+          "--model NAME",
+          "Model to evaluate (will eval all models if not specified)",
+        ) { |model| options.model = model }
+
+        opts.on("-l", "--list", "List evals") { |model| options.list = true }
+      end
+
+    show_help = ARGV.empty?
+    parser.parse!
+
+    if show_help
+      puts parser
+      exit 0
+    end
+
+    options
+  end
+end
diff --git a/evals/lib/eval.rb b/evals/lib/eval.rb
new file mode 100644
index 00000000..b5db9427
--- /dev/null
+++ b/evals/lib/eval.rb
@@ -0,0 +1,136 @@
+#frozen_string_literal: true
+
+class DiscourseAi::Evals::Eval
+  attr_reader :type,
+              :path,
+              :name,
+              :description,
+              :id,
+              :args,
+              :vision,
+              :expected_output,
+              :expected_output_regex
+
+  def initialize(path:)
+    @yaml = YAML.load_file(path).symbolize_keys
+    @path = path
+    @name = @yaml[:name]
+    @id = @yaml[:id]
+    @description = @yaml[:description]
+    @vision = @yaml[:vision]
+    @args = @yaml[:args]&.symbolize_keys
+    @type = @yaml[:type]
+    @expected_output = @yaml[:expected_output]
+    @expected_output_regex = @yaml[:expected_output_regex]
+    @expected_output_regex =
+      Regexp.new(@expected_output_regex, Regexp::MULTILINE) if @expected_output_regex
+
+    @args[:path] = File.expand_path(File.join(File.dirname(path), @args[:path])) if @args&.key?(
+      :path,
+    )
+  end
+
+  def run(llm:)
+    result =
+      case type
+      when "helper"
+        helper(llm, **args)
+      when "pdf_to_text"
+        pdf_to_text(llm, **args)
+      when "image_to_text"
+        image_to_text(llm, **args)
+      end
+
+    if expected_output
+      if result == expected_output
+        { result: :pass }
+      else
+        { result: :fail, expected_output: expected_output, actual_output: result }
+      end
+    elsif expected_output_regex
+      if result.match?(expected_output_regex)
+        { result: :pass }
+      else
+        { result: :fail, expected_output: expected_output_regex, actual_output: result }
+      end
+    else
+      { result: :unknown, actual_output: result }
+    end
+  end
+
+  def print
+    puts "#{id}: #{description}"
+  end
+
+  def to_json
+    {
+      type: @type,
+      path: @path,
+      name: @name,
+      description: @description,
+      id: @id,
+      args: @args,
+      vision: @vision,
+      expected_output: @expected_output,
+      expected_output_regex: @expected_output_regex,
+    }.compact
+  end
+
+  private
+
+  def helper(llm, input:, name:)
+    completion_prompt = CompletionPrompt.find_by(name: name)
+    helper = DiscourseAi::AiHelper::Assistant.new(helper_llm: llm.llm_proxy)
+    result =
+      helper.generate_and_send_prompt(
+        completion_prompt,
+        input,
+        current_user = Discourse.system_user,
+        _force_default_locale = false,
+      )
+
+    result[:suggestions].first
+  end
+
+  def image_to_text(llm, path:)
+    upload =
+      UploadCreator.new(File.open(path), File.basename(path)).create_for(Discourse.system_user.id)
+
+    text = +""
+    DiscourseAi::Utils::ImageToText
+      .new(upload: upload, llm_model: llm.llm_model, user: Discourse.system_user)
+      .extract_text do |chunk, error|
+        text << chunk if chunk
+        text << "\n\n" if chunk
+      end
+    text
+  ensure
+    upload.destroy if upload
+  end
+
+  def pdf_to_text(llm, path:)
+    upload =
+      UploadCreator.new(File.open(path), File.basename(path)).create_for(Discourse.system_user.id)
+
+    uploads =
+      DiscourseAi::Utils::PdfToImages.new(
+        upload: upload,
+        user: Discourse.system_user,
+      ).uploaded_pages
+
+    text = +""
+    uploads.each do |page_upload|
+      DiscourseAi::Utils::ImageToText
+        .new(upload: page_upload, llm_model: llm.llm_model, user: Discourse.system_user)
+        .extract_text do |chunk, error|
+          text << chunk if chunk
+          text << "\n\n" if chunk
+        end
+      upload.destroy
+    end
+
+    text
+  ensure
+    upload.destroy if upload
+  end
+end
diff --git a/evals/lib/llm.rb b/evals/lib/llm.rb
new file mode 100644
index 00000000..1e31a50a
--- /dev/null
+++ b/evals/lib/llm.rb
@@ -0,0 +1,129 @@
+# frozen_string_literal: true
+
+class DiscourseAi::Evals::Llm
+  CONFIGS = {
+    "gpt-4o" => {
+      display_name: "GPT-4o",
+      name: "gpt-4o",
+      tokenizer: "DiscourseAi::Tokenizer::OpenAiTokenizer",
+      api_key_env: "OPENAI_API_KEY",
+      provider: "open_ai",
+      url: "https://api.openai.com/v1/chat/completions",
+      max_prompt_tokens: 131_072,
+      vision_enabled: true,
+    },
+    "gpt-4o-mini" => {
+      display_name: "GPT-4o-mini",
+      name: "gpt-4o-mini",
+      tokenizer: "DiscourseAi::Tokenizer::OpenAiTokenizer",
+      api_key_env: "OPENAI_API_KEY",
+      provider: "open_ai",
+      url: "https://api.openai.com/v1/chat/completions",
+      max_prompt_tokens: 131_072,
+      vision_enabled: true,
+    },
+    "claude-3.5-haiku" => {
+      display_name: "Claude 3.5 Haiku",
+      name: "claude-3-5-haiku-latest",
+      tokenizer: "DiscourseAi::Tokenizer::AnthropicTokenizer",
+      api_key_env: "ANTHROPIC_API_KEY",
+      provider: "anthropic",
+      url: "https://api.anthropic.com/v1/messages",
+      max_prompt_tokens: 200_000,
+      vision_enabled: false,
+    },
+    "claude-3.5-sonnet" => {
+      display_name: "Claude 3.5 Sonnet",
+      name: "claude-3-5-sonnet-latest",
+      tokenizer: "DiscourseAi::Tokenizer::AnthropicTokenizer",
+      api_key_env: "ANTHROPIC_API_KEY",
+      provider: "anthropic",
+      url: "https://api.anthropic.com/v1/messages",
+      max_prompt_tokens: 200_000,
+      vision_enabled: true,
+    },
+    "gemini-2.0-flash" => {
+      display_name: "Gemini 2.0 Flash",
+      name: "gemini-2-0-flash",
+      tokenizer: "DiscourseAi::Tokenizer::GeminiTokenizer",
+      api_key_env: "GEMINI_API_KEY",
+      provider: "google",
+      url: "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash",
+      max_prompt_tokens: 1_000_000,
+      vision_enabled: true,
+    },
+    "gemini-2.0-pro-exp" => {
+      display_name: "Gemini 2.0 pro",
+      name: "gemini-2-0-pro-exp",
+      tokenizer: "DiscourseAi::Tokenizer::GeminiTokenizer",
+      api_key_env: "GEMINI_API_KEY",
+      provider: "google",
+      url: "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-pro-exp",
+      max_prompt_tokens: 1_000_000,
+      vision_enabled: true,
+    },
+  }
+
+  def self.print
+    CONFIGS
+      .keys
+      .map do |config_name|
+        begin
+          new(config_name)
+        rescue StandardError
+          nil
+        end
+      end
+      .compact
+      .each { |llm| puts "#{llm.config_name}: #{llm.name} (#{llm.provider})" }
+  end
+
+  def self.choose(config_name)
+    if CONFIGS[config_name].nil?
+      CONFIGS
+        .keys
+        .map do |config_name|
+          begin
+            new(config_name)
+          rescue => e
+            puts "Error initializing #{config_name}: #{e}"
+            nil
+          end
+        end
+        .compact
+    elsif !CONFIGS.include?(config_name)
+      raise "Invalid llm"
+    else
+      [new(config_name)]
+    end
+  end
+
+  attr_reader :llm_model
+  attr_reader :llm_proxy
+  attr_reader :config_name
+
+  def initialize(config_name)
+    config = CONFIGS[config_name].dup
+    api_key_env = config.delete(:api_key_env)
+    if !ENV[api_key_env]
+      raise "Missing API key for #{config_name}, should be set via #{api_key_env}"
+    end
+
+    config[:api_key] = ENV[api_key_env]
+    @llm_model = LlmModel.new(config)
+    @llm_proxy = DiscourseAi::Completions::Llm.proxy(@llm_model)
+    @config_name = config_name
+  end
+
+  def provider
+    @llm_model.provider
+  end
+
+  def name
+    @llm_model.display_name
+  end
+
+  def vision?
+    @llm_model.vision_enabled
+  end
+end
diff --git a/evals/lib/runner.rb b/evals/lib/runner.rb
new file mode 100644
index 00000000..87cc153e
--- /dev/null
+++ b/evals/lib/runner.rb
@@ -0,0 +1,183 @@
+#frozen_string_literal: true
+
+class DiscourseAi::Evals::Runner
+  class StructuredLogger
+    def initialize
+      @log = []
+      @current_step = @log
+    end
+
+    def log(name, args: nil, start_time: nil, end_time: nil)
+      start_time ||= Time.now.utc
+      end_time ||= Time.now.utc
+      args ||= {}
+      object = { name: name, args: args, start_time: start_time, end_time: end_time }
+      @current_step << object
+    end
+
+    def step(name, args: nil)
+      start_time = Time.now.utc
+      start_step = @current_step
+
+      new_step = { type: :step, name: name, args: args || {}, log: [], start_time: start_time }
+
+      @current_step << new_step
+      @current_step = new_step[:log]
+      yield new_step
+      @current_step = start_step
+      new_step[:end_time] = Time.now.utc
+    end
+
+    def to_trace_event_json
+      trace_events = []
+      process_id = 1
+      thread_id = 1
+
+      to_trace_event(@log, process_id, thread_id, trace_events)
+
+      JSON.pretty_generate({ traceEvents: trace_events })
+    end
+
+    private
+
+    def to_trace_event(log_items, pid, tid, trace_events, parent_start_time = nil)
+      log_items.each do |item|
+        if item.is_a?(Hash) && item[:type] == :step
+          trace_events << {
+            name: item[:name],
+            cat: "default",
+            ph: "B", # Begin event
+            pid: pid,
+            tid: tid,
+            args: item[:args],
+            ts: timestamp_in_microseconds(item[:start_time]),
+          }
+
+          to_trace_event(item[:log], pid, tid, trace_events, item[:start_time])
+
+          trace_events << {
+            name: item[:name],
+            cat: "default",
+            ph: "E", # End event
+            pid: pid,
+            tid: tid,
+            ts: timestamp_in_microseconds(item[:end_time]),
+          }
+        else
+          trace_events << {
+            name: item[:name],
+            cat: "default",
+            ph: "B",
+            pid: pid,
+            tid: tid,
+            args: item[:args],
+            ts: timestamp_in_microseconds(item[:start_time] || parent_start_time || Time.now.utc),
+            s: "p", # Scope: process
+          }
+          trace_events << {
+            name: item[:name],
+            cat: "default",
+            ph: "E",
+            pid: pid,
+            tid: tid,
+            ts: timestamp_in_microseconds(item[:end_time] || Time.now.utc),
+            s: "p",
+          }
+        end
+      end
+    end
+
+    def timestamp_in_microseconds(time)
+      (time.to_f * 1_000_000).to_i
+    end
+  end
+
+  attr_reader :llms, :cases
+
+  def self.evals_paths
+    @eval_paths ||= Dir.glob(File.join(File.join(__dir__, "../cases"), "*/*.yml"))
+  end
+
+  def self.evals
+    @evals ||= evals_paths.map { |path| DiscourseAi::Evals::Eval.new(path: path) }
+  end
+
+  def self.print
+    evals.each(&:print)
+  end
+
+  def initialize(eval_name:, llms:)
+    @llms = llms
+    @eval = self.class.evals.find { |c| c.id == eval_name }
+
+    if !@eval
+      puts "Error: Unknown evaluation '#{eval_name}'"
+      exit 1
+    end
+
+    if @llms.empty?
+      puts "Error: Unknown model 'model'"
+      exit 1
+    end
+  end
+
+  def run!
+    puts "Running evaluation '#{@eval.id}'"
+
+    structured_log_filename = "#{@eval.id}-#{Time.now.strftime("%Y%m%d-%H%M%S")}.json"
+    log_filename = "#{@eval.id}-#{Time.now.strftime("%Y%m%d-%H%M%S")}.log"
+    logs_dir = File.join(__dir__, "../log")
+    FileUtils.mkdir_p(logs_dir)
+
+    log_path = File.expand_path(File.join(logs_dir, log_filename))
+    structured_log_path = File.expand_path(File.join(logs_dir, structured_log_filename))
+
+    logger = Logger.new(File.open(log_path, "a"))
+    logger.info("Starting evaluation '#{@eval.id}'")
+
+    Thread.current[:llm_audit_log] = logger
+    structured_logger = Thread.current[:llm_audit_structured_log] = StructuredLogger.new
+
+    structured_logger.step("Evaluating #{@eval.id}", args: @eval.to_json) do
+      llms.each do |llm|
+        if @eval.vision && !llm.vision?
+          logger.info("Skipping LLM: #{llm.name} as it does not support vision")
+          next
+        end
+
+        structured_logger.step("Evaluating with LLM: #{llm.name}") do |step|
+          logger.info("Evaluating with LLM: #{llm.name}")
+          print "#{llm.name}: "
+          result = @eval.run(llm: llm)
+
+          step[:args] = result
+          step[:cname] = result[:result] == :pass ? :good : :bad
+
+          if result[:result] == :fail
+            puts "Failed 🔴"
+            puts "---- Expected ----\n#{result[:expected_output]}"
+            puts "---- Actual ----\n#{result[:actual_output]}"
+            logger.error("Evaluation failed with LLM: #{llm.name}")
+          elsif result[:result] == :pass
+            puts "Passed 🟢"
+            logger.info("Evaluation passed with LLM: #{llm.name}")
+          else
+            STDERR.puts "Error: Unknown result #{eval.inspect}"
+            logger.error("Unknown result: #{eval.inspect}")
+          end
+        end
+      end
+    end
+
+    #structured_logger.save(structured_log_path)
+
+    File.write("#{structured_log_path}", structured_logger.to_trace_event_json)
+
+    puts
+    puts "Log file: #{log_path}"
+    puts "Structured log file (ui.perfetto.dev): #{structured_log_path}"
+
+    # temp code
+    # puts File.read(structured_log_path)
+  end
+end
diff --git a/evals/run b/evals/run
new file mode 100755
index 00000000..a4a39697
--- /dev/null
+++ b/evals/run
@@ -0,0 +1,25 @@
+#!/usr/bin/env ruby
+# frozen_string_literal: true
+
+require_relative "lib/boot"
+require_relative "lib/llm"
+require_relative "lib/cli"
+require_relative "lib/runner"
+require_relative "lib/eval"
+
+options = DiscourseAi::Evals::Cli.parse_options!
+
+if options.list
+  DiscourseAi::Evals::Runner.print
+  exit 0
+end
+
+if options.list_models
+  DiscourseAi::Evals::Llm.print
+  exit 0
+end
+
+DiscourseAi::Evals::Runner.new(
+  eval_name: options.eval_name,
+  llms: DiscourseAi::Evals::Llm.choose(options.model),
+).run!
diff --git a/lib/ai_bot/bot.rb b/lib/ai_bot/bot.rb
index f2b1d53c..d3e5f73c 100644
--- a/lib/ai_bot/bot.rb
+++ b/lib/ai_bot/bot.rb
@@ -16,7 +16,8 @@ module DiscourseAi
       def initialize(bot_user, persona, model = nil)
         @bot_user = bot_user
         @persona = persona
-        @model = model || self.class.guess_model(bot_user) || @persona.class.default_llm
+        @model =
+          model || self.class.guess_model(bot_user) || LlmModel.find(@persona.class.default_llm_id)
       end
 
       attr_reader :bot_user
@@ -245,7 +246,7 @@ module DiscourseAi
 
         return if associated_llm.nil? # Might be a persona user. Handled by constructor.
 
-        "custom:#{associated_llm.id}"
+        associated_llm
       end
 
       def build_placeholder(summary, details, custom_raw: nil)
diff --git a/lib/ai_bot/personas/persona.rb b/lib/ai_bot/personas/persona.rb
index 58b05e26..33c2907e 100644
--- a/lib/ai_bot/personas/persona.rb
+++ b/lib/ai_bot/personas/persona.rb
@@ -17,7 +17,7 @@ module DiscourseAi
             1_048_576
           end
 
-          def question_consolidator_llm
+          def question_consolidator_llm_id
             nil
           end
 
@@ -173,9 +173,11 @@ module DiscourseAi
           TEXT
 
           question_consolidator_llm = llm
-          if self.class.question_consolidator_llm.present?
-            question_consolidator_llm =
-              DiscourseAi::Completions::Llm.proxy(self.class.question_consolidator_llm)
+          if self.class.question_consolidator_llm_id.present?
+            question_consolidator_llm ||=
+              DiscourseAi::Completions::Llm.proxy(
+                LlmModel.find_by(id: self.class.question_consolidator_llm_id),
+              )
           end
 
           if context[:custom_instructions].present?
diff --git a/lib/ai_helper/assistant.rb b/lib/ai_helper/assistant.rb
index 7333db35..8de1bcf8 100644
--- a/lib/ai_helper/assistant.rb
+++ b/lib/ai_helper/assistant.rb
@@ -13,6 +13,20 @@ module DiscourseAi
         prompt_cache.flush!
       end
 
+      def initialize(helper_llm: nil, image_caption_llm: nil)
+        @helper_llm = helper_llm
+        @image_caption_llm = image_caption_llm
+      end
+
+      def helper_llm
+        @helper_llm || DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_helper_model)
+      end
+
+      def image_caption_llm
+        @image_caption_llm ||
+          DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_helper_image_caption_model)
+      end
+
       def available_prompts(user)
         key = "prompt_cache_#{I18n.locale}"
         self
@@ -115,7 +129,7 @@ module DiscourseAi
       end
 
       def generate_prompt(completion_prompt, input, user, force_default_locale = false, &block)
-        llm = DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_helper_model)
+        llm = helper_llm
         prompt = completion_prompt.messages_with_input(input)
         localize_prompt!(prompt, user, force_default_locale)
 
@@ -182,7 +196,7 @@ module DiscourseAi
           )
 
         raw_caption =
-          DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_helper_image_caption_model).generate(
+          image_caption_llm.generate(
             prompt,
             user: user,
             max_tokens: 1024,
diff --git a/lib/completions/endpoints/base.rb b/lib/completions/endpoints/base.rb
index a4b6050f..74933f10 100644
--- a/lib/completions/endpoints/base.rb
+++ b/lib/completions/endpoints/base.rb
@@ -156,7 +156,7 @@ module DiscourseAi
                 )
 
               if !@streaming_mode
-                return(
+                response_data =
                   non_streaming_response(
                     response: response,
                     xml_tool_processor: xml_tool_processor,
@@ -164,7 +164,7 @@ module DiscourseAi
                     partials_raw: partials_raw,
                     response_raw: response_raw,
                   )
-                )
+                return response_data
               end
 
               begin
@@ -223,10 +223,46 @@ module DiscourseAi
                 log.duration_msecs = (Time.now - start_time) * 1000
                 log.save!
                 LlmQuota.log_usage(@llm_model, user, log.request_tokens, log.response_tokens)
-                if Rails.env.development?
+                if Rails.env.development? && !ENV["DISCOURSE_AI_NO_DEBUG"]
                   puts "#{self.class.name}: request_tokens #{log.request_tokens} response_tokens #{log.response_tokens}"
                 end
               end
+              if log && (logger = Thread.current[:llm_audit_log])
+                call_data = <<~LOG
+                  #{self.class.name}: request_tokens #{log.request_tokens} response_tokens #{log.response_tokens}
+                  request:
+                  #{format_possible_json_payload(log.raw_request_payload)}
+                  response:
+                  #{response_data}
+                LOG
+                logger.info(call_data)
+              end
+              if log && (structured_logger = Thread.current[:llm_audit_structured_log])
+                llm_request =
+                  begin
+                    JSON.parse(log.raw_request_payload)
+                  rescue StandardError
+                    log.raw_request_payload
+                  end
+
+                # gemini puts passwords in query params
+                # we don't want to log that
+                structured_logger.log(
+                  "llm_call",
+                  args: {
+                    class: self.class.name,
+                    completion_url: request.uri.to_s.split("?")[0],
+                    request: llm_request,
+                    result: response_data,
+                    request_tokens: log.request_tokens,
+                    response_tokens: log.response_tokens,
+                    duration: log.duration_msecs,
+                    stream: @streaming_mode,
+                  },
+                  start_time: start_time.utc,
+                  end_time: Time.now.utc,
+                )
+              end
             end
           end
         end
@@ -298,6 +334,14 @@ module DiscourseAi
 
         private
 
+        def format_possible_json_payload(payload)
+          begin
+            JSON.pretty_generate(JSON.parse(payload))
+          rescue JSON::ParserError
+            payload
+          end
+        end
+
         def start_log(
           provider_id:,
           request_body:,
diff --git a/lib/configuration/llm_enumerator.rb b/lib/configuration/llm_enumerator.rb
index 1209001d..f7de3e7c 100644
--- a/lib/configuration/llm_enumerator.rb
+++ b/lib/configuration/llm_enumerator.rb
@@ -16,11 +16,8 @@ module DiscourseAi
 
           AiPersona
             .where("force_default_llm = ?", true)
-            .pluck(:default_llm, :name, :id)
-            .each do |llm_name, name, id|
-              llm_id = llm_name.split(":").last.to_i
-              rval[llm_id] << { type: :ai_persona, name: name, id: id }
-            end
+            .pluck(:default_llm_id, :name, :id)
+            .each { |llm_id, name, id| rval[llm_id] << { type: :ai_persona, name: name, id: id } }
         end
 
         if SiteSetting.ai_helper_enabled
@@ -50,6 +47,26 @@ module DiscourseAi
         true
       end
 
+      # returns an array of hashes (id: , name:, vision_enabled:)
+      def self.values_for_serialization(allowed_seeded_llm_ids: nil)
+        builder = DB.build(<<~SQL)
+          SELECT id, display_name AS name, vision_enabled
+          FROM llm_models
+          /*where*/
+        SQL
+
+        if allowed_seeded_llm_ids.is_a?(Array) && !allowed_seeded_llm_ids.empty?
+          builder.where(
+            "id > 0 OR id IN (:allowed_seeded_llm_ids)",
+            allowed_seeded_llm_ids: allowed_seeded_llm_ids,
+          )
+        else
+          builder.where("id > 0")
+        end
+
+        builder.query_hash.map(&:symbolize_keys)
+      end
+
       def self.values(allowed_seeded_llms: nil)
         values = DB.query_hash(<<~SQL).map(&:symbolize_keys)
           SELECT display_name AS name, id AS value
diff --git a/lib/discord/bot/persona_replier.rb b/lib/discord/bot/persona_replier.rb
index c0f51d24..ad9a7841 100644
--- a/lib/discord/bot/persona_replier.rb
+++ b/lib/discord/bot/persona_replier.rb
@@ -13,7 +13,7 @@ module DiscourseAi
           DiscourseAi::AiBot::Bot.as(
             Discourse.system_user,
             persona: @persona,
-            model: @persona.class.default_llm,
+            model: LlmModel.find(@persona.class.default_llm_id),
           )
         super(body)
       end
diff --git a/lib/utils/image_to_text.rb b/lib/utils/image_to_text.rb
new file mode 100644
index 00000000..43f3c671
--- /dev/null
+++ b/lib/utils/image_to_text.rb
@@ -0,0 +1,213 @@
+# frozen_string_literal: true
+
+class DiscourseAi::Utils::ImageToText
+  BACKOFF_SECONDS = [5, 30, 60]
+  MAX_IMAGE_SIZE = 10.megabytes
+
+  class Reader
+    def initialize(uploads:, llm_model:, user:)
+      @uploads = uploads
+      @llm_model = llm_model
+      @user = user
+      @buffer = +""
+
+      @to_process = uploads.dup
+    end
+
+    # return nil if no more data
+    def read(length)
+      # for implementation simplicity we will process one image at a time
+      if !@buffer.empty?
+        part = @buffer.slice!(0, length)
+        return part
+      end
+
+      return nil if @to_process.empty?
+
+      upload = @to_process.shift
+      extractor =
+        DiscourseAi::Utils::ImageToText.new(upload: upload, llm_model: @llm_model, user: @user)
+      extractor.extract_text do |chunk, error|
+        if error
+          Discourse.warn_exception(
+            error,
+            message: "Discourse AI: Failed to extract text from image",
+          )
+        else
+          # this introduces chunk markers so discourse rag ingestion requires no overlaps
+          @buffer << "\n[[metadata ]]\n"
+          @buffer << chunk
+        end
+      end
+
+      read(length)
+    end
+  end
+
+  def self.as_fake_file(uploads:, llm_model:, user:)
+    # given our implementation for extracting text expect a file, return a simple object that can simulate read(size)
+    # and stream content
+    Reader.new(uploads: uploads, llm_model: llm_model, user: user)
+  end
+
+  attr_reader :upload, :llm_model, :user
+
+  def initialize(upload:, llm_model:, user:)
+    @upload = upload
+    @llm_model = llm_model
+    @user = user
+  end
+
+  def extract_text(retries: 3)
+    uploads ||= @uploaded_pages
+
+    raise "must specify a block" if !block_given?
+    extracted = nil
+    error = nil
+
+    backoff = BACKOFF_SECONDS.dup
+
+    retries.times do
+      seconds = nil
+      begin
+        extracted = extract_text_from_page(upload)
+        break
+      rescue => e
+        error = e
+        seconds = backoff.shift || seconds
+        sleep(seconds)
+      end
+    end
+    if extracted
+      extracted.each { |chunk| yield(chunk) }
+    else
+      yield(nil, error)
+    end
+    extracted || []
+  end
+
+  private
+
+  def system_message
+    <<~MSG
+      OCR the following page into Markdown. Tables should be formatted as Github flavored markdown.
+      Do not surround your output with triple backticks.
+
+      Chunk the document into sections of roughly 250 - 1000 words. Our goal is to identify parts of the page with same semantic theme. These chunks will be embedded and used in a RAG pipeline.
+
+      Always prefer returning text in Markdown vs HTML.
+      Describe all the images and graphs you encounter.
+      Only return text that will assist in the querying of data. Omit text such as "I had trouble recognizing images" and so on.
+
+      Surround the chunks with <chunk> </chunk> html tags.
+    MSG
+  end
+
+  def extract_text_from_page(page)
+    raw_text = extract_text_with_tesseract(page)
+
+    llm = llm_model.to_llm
+    if raw_text.present?
+      messages = [
+        {
+          type: :user,
+          content:
+            "The following text was extracted from an image using OCR. Please enhance, correct, and structure this content while maintaining the original meaning:\n\n#{raw_text}",
+          upload_ids: [page.id],
+        },
+      ]
+    else
+      messages = [
+        { type: :user, content: "Please OCR the content in the image.", upload_ids: [page.id] },
+      ]
+    end
+    prompt = DiscourseAi::Completions::Prompt.new(system_message, messages: messages)
+    result = llm.generate(prompt, user: Discourse.system_user)
+    extract_chunks(result)
+  end
+
+  def extract_text_with_tesseract(page)
+    upload_path =
+      if page.local?
+        Discourse.store.path_for(page)
+      else
+        Discourse.store.download_safe(page, max_file_size_kb: MAX_IMAGE_SIZE)&.path
+      end
+
+    return "" if !upload_path || !File.exist?(upload_path)
+
+    tmp_output_file = Tempfile.new(%w[tesseract_output .txt])
+    tmp_output = tmp_output_file.path
+    tmp_output_file.unlink
+
+    command = [
+      "tesseract",
+      upload_path,
+      tmp_output.sub(/\.txt$/, ""), # Tesseract adds .txt automatically
+    ]
+
+    success =
+      Discourse::Utils.execute_command(
+        *command,
+        timeout: 20.seconds,
+        failure_message: "Failed to OCR image with Tesseract",
+      )
+
+    if success && File.exist?("#{tmp_output}")
+      text = File.read("#{tmp_output}")
+      begin
+        File.delete("#{tmp_output}")
+      rescue StandardError
+        nil
+      end
+      text.strip
+    else
+      Rails.logger.error("Tesseract OCR failed for #{upload_path}")
+      ""
+    end
+  rescue => e
+    Rails.logger.error("Error during OCR processing: #{e.message}")
+    ""
+  end
+
+  def extract_chunks(text)
+    return [] if text.nil? || text.empty?
+
+    if text.include?("<chunk>") && text.include?("</chunk>")
+      chunks = []
+      remaining_text = text.dup
+
+      while remaining_text.length > 0
+        if remaining_text.start_with?("<chunk>")
+          # Extract chunk content
+          chunk_end = remaining_text.index("</chunk>")
+          if chunk_end
+            chunk = remaining_text[7..chunk_end - 1].strip
+            chunks << chunk unless chunk.empty?
+            remaining_text = remaining_text[chunk_end + 8..-1] || ""
+          else
+            # Malformed chunk - add remaining text and break
+            chunks << remaining_text[7..-1].strip
+            break
+          end
+        else
+          # Handle text before next chunk if it exists
+          next_chunk = remaining_text.index("<chunk>")
+          if next_chunk
+            text_before = remaining_text[0...next_chunk].strip
+            chunks << text_before unless text_before.empty?
+            remaining_text = remaining_text[next_chunk..-1]
+          else
+            # No more chunks - add remaining text and break
+            chunks << remaining_text.strip
+            break
+          end
+        end
+      end
+
+      return chunks.reject(&:empty?)
+    end
+
+    [text]
+  end
+end
diff --git a/lib/utils/pdf_to_images.rb b/lib/utils/pdf_to_images.rb
new file mode 100644
index 00000000..b9865344
--- /dev/null
+++ b/lib/utils/pdf_to_images.rb
@@ -0,0 +1,80 @@
+# frozen_string_literal: true
+
+class DiscourseAi::Utils::PdfToImages
+  MAX_PDF_SIZE = 100.megabytes
+  # this is long, mutool can be faster than magick, 10 minutes will be enough for quite large pdfs
+  MAX_CONVERT_SECONDS = 600
+  BACKOFF_SECONDS = [5, 30, 60]
+
+  attr_reader :upload, :user
+
+  def initialize(upload:, user:)
+    @upload = upload
+    @user = user
+    @uploaded_pages = UploadReference.where(target: upload).map(&:upload).presence
+  end
+
+  def uploaded_pages
+    @uploaded_pages ||= extract_pages
+  end
+
+  def extract_pages
+    Dir.mktmpdir("discourse-pdf-#{SecureRandom.hex(8)}")
+
+    begin
+      pdf_path =
+        if upload.local?
+          Discourse.store.path_for(upload)
+        else
+          Discourse.store.download_safe(upload, max_file_size_kb: MAX_PDF_SIZE)&.path
+        end
+
+      raise Discourse::InvalidParameters.new("Failed to download PDF") if pdf_path.nil?
+
+      temp_pdf = File.join(temp_dir, "source.pdf")
+      FileUtils.cp(pdf_path, temp_pdf)
+
+      # Convert PDF to individual page images
+      output_pattern = File.join(temp_dir, "page-%04d.png")
+
+      command = [
+        "magick",
+        "-density",
+        "300",
+        temp_pdf,
+        "-background",
+        "white",
+        "-auto-orient",
+        "-quality",
+        "85",
+        output_pattern,
+      ]
+
+      Discourse::Utils.execute_command(
+        *command,
+        failure_message: "Failed to convert PDF to images",
+        timeout: MAX_CONVERT_SECONDS,
+      )
+
+      uploads = []
+      Dir
+        .glob(File.join(temp_dir, "page-*.png"))
+        .sort
+        .each do |page_path|
+          upload =
+            UploadCreator.new(File.open(page_path), "page-#{File.basename(page_path)}").create_for(
+              @user.id,
+            )
+
+          uploads << upload
+        end
+
+      # Create upload references
+      UploadReference.ensure_exist!(upload_ids: uploads.map(&:id), target: @upload)
+
+      @uploaded_pages = uploads
+    ensure
+      FileUtils.rm_rf(temp_dir)
+    end
+  end
+end
diff --git a/spec/configuration/llm_enumerator_spec.rb b/spec/configuration/llm_enumerator_spec.rb
index 771dff35..daa6b441 100644
--- a/spec/configuration/llm_enumerator_spec.rb
+++ b/spec/configuration/llm_enumerator_spec.rb
@@ -2,6 +2,39 @@
 
 RSpec.describe DiscourseAi::Configuration::LlmEnumerator do
   fab!(:fake_model)
+  fab!(:llm_model)
+  fab!(:seeded_model)
+
+  describe "#values_for_serialization" do
+    it "returns an array for that can be used for serialization" do
+      fake_model.destroy!
+
+      expect(described_class.values_for_serialization).to eq(
+        [
+          {
+            id: llm_model.id,
+            name: llm_model.display_name,
+            vision_enabled: llm_model.vision_enabled,
+          },
+        ],
+      )
+
+      expect(
+        described_class.values_for_serialization(allowed_seeded_llm_ids: [seeded_model.id.to_s]),
+      ).to contain_exactly(
+        {
+          id: seeded_model.id,
+          name: seeded_model.display_name,
+          vision_enabled: seeded_model.vision_enabled,
+        },
+        {
+          id: llm_model.id,
+          name: llm_model.display_name,
+          vision_enabled: llm_model.vision_enabled,
+        },
+      )
+    end
+  end
 
   describe "#global_usage" do
     before do
diff --git a/spec/jobs/regular/digest_rag_upload_spec.rb b/spec/jobs/regular/digest_rag_upload_spec.rb
index 97da3b79..eec30ac6 100644
--- a/spec/jobs/regular/digest_rag_upload_spec.rb
+++ b/spec/jobs/regular/digest_rag_upload_spec.rb
@@ -2,8 +2,8 @@
 
 RSpec.describe Jobs::DigestRagUpload do
   fab!(:persona) { Fabricate(:ai_persona) }
-  fab!(:upload)
-
+  fab!(:upload) { Fabricate(:upload, extension: "txt") }
+  fab!(:pdf_upload) { Fabricate(:upload, extension: "pdf") }
   let(:document_file) { StringIO.new("some text" * 200) }
 
   fab!(:cloudflare_embedding_def)
@@ -31,6 +31,19 @@ RSpec.describe Jobs::DigestRagUpload do
   end
 
   describe "#execute" do
+    context "when processing a PDF upload" do
+      it "will reject the indexing if the site setting is not enabled" do
+        SiteSetting.ai_rag_pdf_images_enabled = false
+
+        expect {
+          described_class.new.execute(
+            upload_id: pdf_upload.id,
+            target_id: persona.id,
+            target_type: persona.class.to_s,
+          )
+        }.to raise_error(Discourse::InvalidAccess)
+      end
+    end
     context "when processing an upload containing metadata" do
       it "correctly splits on metadata boundary" do
         # be explicit here about chunking strategy
diff --git a/spec/jobs/regular/stream_discord_reply_spec.rb b/spec/jobs/regular/stream_discord_reply_spec.rb
index aa468b75..c91f1e38 100644
--- a/spec/jobs/regular/stream_discord_reply_spec.rb
+++ b/spec/jobs/regular/stream_discord_reply_spec.rb
@@ -13,9 +13,12 @@ RSpec.describe Jobs::StreamDiscordReply, type: :job do
     }.to_json.to_s
   end
 
+  fab!(:llm_model)
+  fab!(:persona) { Fabricate(:ai_persona, default_llm_id: llm_model.id) }
+
   before do
     SiteSetting.ai_discord_search_mode = "persona"
-    SiteSetting.ai_discord_search_persona = -1
+    SiteSetting.ai_discord_search_persona = persona.id
   end
 
   it "calls PersonaReplier when search mode is persona" do
diff --git a/spec/lib/discord/bot/persona_replier_spec.rb b/spec/lib/discord/bot/persona_replier_spec.rb
index 53050cf7..00bb4859 100644
--- a/spec/lib/discord/bot/persona_replier_spec.rb
+++ b/spec/lib/discord/bot/persona_replier_spec.rb
@@ -8,8 +8,11 @@ RSpec.describe DiscourseAi::Discord::Bot::PersonaReplier do
   end
   let(:persona_replier) { described_class.new(interaction_body) }
 
+  fab!(:llm_model)
+  fab!(:persona) { Fabricate(:ai_persona, default_llm_id: llm_model.id) }
+
   before do
-    SiteSetting.ai_discord_search_persona = "-1"
+    SiteSetting.ai_discord_search_persona = persona.id.to_s
     allow_any_instance_of(DiscourseAi::AiBot::Bot).to receive(:reply).and_return(
       "This is a reply from bot!",
     )
diff --git a/spec/lib/modules/ai_bot/entry_point_spec.rb b/spec/lib/modules/ai_bot/entry_point_spec.rb
index 4f443280..57da71c3 100644
--- a/spec/lib/modules/ai_bot/entry_point_spec.rb
+++ b/spec/lib/modules/ai_bot/entry_point_spec.rb
@@ -43,7 +43,7 @@ RSpec.describe DiscourseAi::AiBot::EntryPoint do
             :ai_persona,
             enabled: true,
             allowed_group_ids: [bot_allowed_group.id],
-            default_llm: "claude-2",
+            default_llm_id: claude_2.id,
             force_default_llm: true,
           )
         persona.create_user!
diff --git a/spec/lib/modules/ai_bot/personas/persona_spec.rb b/spec/lib/modules/ai_bot/personas/persona_spec.rb
index 34d056a8..8949ae13 100644
--- a/spec/lib/modules/ai_bot/personas/persona_spec.rb
+++ b/spec/lib/modules/ai_bot/personas/persona_spec.rb
@@ -8,7 +8,6 @@ class TestPersona < DiscourseAi::AiBot::Personas::Persona
       DiscourseAi::AiBot::Tools::Image,
     ]
   end
-
   def system_prompt
     <<~PROMPT
       {site_url}
@@ -337,7 +336,7 @@ RSpec.describe DiscourseAi::AiBot::Personas::Persona do
             name: "custom",
             rag_conversation_chunks: 3,
             allowed_group_ids: [Group::AUTO_GROUPS[:trust_level_0]],
-            question_consolidator_llm: "custom:#{llm_model.id}",
+            question_consolidator_llm_id: llm_model.id,
           )
 
         UploadReference.ensure_exist!(target: custom_ai_persona, upload_ids: [upload.id])
diff --git a/spec/lib/modules/ai_bot/playground_spec.rb b/spec/lib/modules/ai_bot/playground_spec.rb
index 4a3e541c..124582ed 100644
--- a/spec/lib/modules/ai_bot/playground_spec.rb
+++ b/spec/lib/modules/ai_bot/playground_spec.rb
@@ -241,7 +241,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
         system_prompt: "You are a helpful bot",
         vision_enabled: true,
         vision_max_pixels: 1_000,
-        default_llm: "custom:#{opus_model.id}",
+        default_llm_id: opus_model.id,
         allow_topic_mentions: true,
       )
     end
@@ -293,7 +293,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
 
       persona.create_user!
       persona.update!(
-        default_llm: "custom:#{claude_2.id}",
+        default_llm_id: claude_2.id,
         allow_chat_channel_mentions: true,
         allow_topic_mentions: true,
       )
@@ -313,7 +313,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
         SiteSetting.ai_bot_enabled = true
         SiteSetting.chat_allowed_groups = "#{Group::AUTO_GROUPS[:trust_level_0]}"
         Group.refresh_automatic_groups!
-        persona.update!(allow_chat_channel_mentions: true, default_llm: "custom:#{opus_model.id}")
+        persona.update!(allow_chat_channel_mentions: true, default_llm_id: opus_model.id)
       end
 
       it "should behave in a sane way when threading is enabled" do
@@ -428,7 +428,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
           allow_chat_direct_messages: true,
           allow_topic_mentions: false,
           allow_chat_channel_mentions: false,
-          default_llm: "custom:#{opus_model.id}",
+          default_llm_id: opus_model.id,
         )
         SiteSetting.ai_bot_enabled = true
       end
@@ -629,7 +629,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
       post = nil
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Yes I can", "Magic Title"],
-        llm: "custom:#{claude_2.id}",
+        llm: claude_2,
       ) do
         post =
           create_post(
@@ -648,10 +648,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
 
       llm2.toggle_companion_user
 
-      DiscourseAi::Completions::Llm.with_prepared_responses(
-        ["Hi from bot two"],
-        llm: "custom:#{llm2.id}",
-      ) do
+      DiscourseAi::Completions::Llm.with_prepared_responses(["Hi from bot two"], llm: llm2) do
         create_post(
           user: admin,
           raw: "hi @#{llm2.user.username.capitalize} how are you",
@@ -664,12 +661,9 @@ RSpec.describe DiscourseAi::AiBot::Playground do
       expect(last_post.user_id).to eq(persona.user_id)
 
       # tether llm, so it can no longer be switched
-      persona.update!(force_default_llm: true, default_llm: "custom:#{claude_2.id}")
+      persona.update!(force_default_llm: true, default_llm_id: claude_2.id)
 
-      DiscourseAi::Completions::Llm.with_prepared_responses(
-        ["Hi from bot one"],
-        llm: "custom:#{claude_2.id}",
-      ) do
+      DiscourseAi::Completions::Llm.with_prepared_responses(["Hi from bot one"], llm: claude_2) do
         create_post(
           user: admin,
           raw: "hi @#{llm2.user.username.capitalize} how are you",
@@ -689,7 +683,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
 
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Yes I can", "Magic Title"],
-        llm: "custom:#{claude_2.id}",
+        llm: claude_2,
       ) do
         post =
           create_post(
@@ -731,11 +725,11 @@ RSpec.describe DiscourseAi::AiBot::Playground do
       toggle_enabled_bots(bots: [gpt_35_turbo, claude_2])
 
       post = nil
-      persona.update!(force_default_llm: true, default_llm: "custom:#{gpt_35_turbo.id}")
+      persona.update!(force_default_llm: true, default_llm_id: gpt_35_turbo.id)
 
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Yes I can", "Magic Title"],
-        llm: "custom:#{gpt_35_turbo.id}",
+        llm: gpt_35_turbo,
       ) do
         post =
           create_post(
@@ -768,7 +762,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
 
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Yes I can", "Magic Title"],
-        llm: "custom:#{gpt_35_turbo.id}",
+        llm: gpt_35_turbo,
       ) do
         messages =
           MessageBus.track_publish do
@@ -805,10 +799,7 @@ RSpec.describe DiscourseAi::AiBot::Playground do
       )
 
       # replies as correct persona if replying direct to persona
-      DiscourseAi::Completions::Llm.with_prepared_responses(
-        ["Another reply"],
-        llm: "custom:#{gpt_35_turbo.id}",
-      ) do
+      DiscourseAi::Completions::Llm.with_prepared_responses(["Another reply"], llm: gpt_35_turbo) do
         create_post(
           raw: "Please ignore this bot, I am replying to a user",
           topic: post.topic,
diff --git a/spec/models/ai_persona_spec.rb b/spec/models/ai_persona_spec.rb
index 2a7e9921..2e3a55de 100644
--- a/spec/models/ai_persona_spec.rb
+++ b/spec/models/ai_persona_spec.rb
@@ -1,6 +1,9 @@
 # frozen_string_literal: true
 
 RSpec.describe AiPersona do
+  fab!(:llm_model)
+  fab!(:seeded_llm_model) { Fabricate(:llm_model, id: -1) }
+
   it "validates context settings" do
     persona =
       AiPersona.new(
@@ -118,7 +121,7 @@ RSpec.describe AiPersona do
     forum_helper = AiPersona.find_by(name: "Forum Helper")
     forum_helper.update!(
       user_id: 1,
-      default_llm: "anthropic:claude-2",
+      default_llm_id: llm_model.id,
       max_context_posts: 3,
       allow_topic_mentions: true,
       allow_personal_messages: true,
@@ -133,7 +136,7 @@ RSpec.describe AiPersona do
     # tl 0 by default
     expect(klass.allowed_group_ids).to eq([10])
     expect(klass.user_id).to eq(1)
-    expect(klass.default_llm).to eq("anthropic:claude-2")
+    expect(klass.default_llm_id).to eq(llm_model.id)
     expect(klass.max_context_posts).to eq(3)
     expect(klass.allow_topic_mentions).to eq(true)
     expect(klass.allow_personal_messages).to eq(true)
@@ -149,7 +152,7 @@ RSpec.describe AiPersona do
         system_prompt: "test",
         tools: [],
         allowed_group_ids: [],
-        default_llm: "anthropic:claude-2",
+        default_llm_id: llm_model.id,
         max_context_posts: 3,
         allow_topic_mentions: true,
         allow_personal_messages: true,
@@ -164,7 +167,7 @@ RSpec.describe AiPersona do
     expect(klass.system).to eq(false)
     expect(klass.allowed_group_ids).to eq([])
     expect(klass.user_id).to eq(1)
-    expect(klass.default_llm).to eq("anthropic:claude-2")
+    expect(klass.default_llm_id).to eq(llm_model.id)
     expect(klass.max_context_posts).to eq(3)
     expect(klass.allow_topic_mentions).to eq(true)
     expect(klass.allow_personal_messages).to eq(true)
@@ -227,10 +230,9 @@ RSpec.describe AiPersona do
         system_prompt: "test",
         tools: [],
         allowed_group_ids: [],
-        default_llm: "seeded_model:-1",
+        default_llm_id: seeded_llm_model.id,
       )
 
-    llm_model = Fabricate(:llm_model, id: -1)
     SiteSetting.ai_bot_allowed_seeded_models = ""
 
     expect(persona.valid?).to eq(false)
diff --git a/spec/requests/admin/ai_llms_controller_spec.rb b/spec/requests/admin/ai_llms_controller_spec.rb
index b8504b92..280204f2 100644
--- a/spec/requests/admin/ai_llms_controller_spec.rb
+++ b/spec/requests/admin/ai_llms_controller_spec.rb
@@ -16,7 +16,7 @@ RSpec.describe DiscourseAi::Admin::AiLlmsController do
         :ai_persona,
         name: "Cool persona",
         force_default_llm: true,
-        default_llm: "custom:#{llm_model2.id}",
+        default_llm_id: llm_model2.id,
       )
     end
 
diff --git a/spec/requests/admin/ai_personas_controller_spec.rb b/spec/requests/admin/ai_personas_controller_spec.rb
index cf991eae..c9c036c0 100644
--- a/spec/requests/admin/ai_personas_controller_spec.rb
+++ b/spec/requests/admin/ai_personas_controller_spec.rb
@@ -4,6 +4,7 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
   fab!(:admin)
   fab!(:ai_persona)
   fab!(:embedding_definition)
+  fab!(:llm_model)
 
   before do
     sign_in(admin)
@@ -27,9 +28,13 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
       expect(response).to be_successful
 
       expect(response.parsed_body["meta"]["llms"]).to eq(
-        DiscourseAi::Configuration::LlmEnumerator.values.map do |hash|
-          { "id" => hash[:value], "name" => hash[:name] }
-        end,
+        [
+          {
+            id: llm_model.id,
+            name: llm_model.display_name,
+            vision_enabled: llm_model.vision_enabled,
+          }.stringify_keys,
+        ],
       )
     end
 
@@ -44,7 +49,8 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
           allow_personal_messages: true,
           allow_chat_channel_mentions: true,
           allow_chat_direct_messages: true,
-          default_llm: "anthropic:claude-2",
+          default_llm_id: llm_model.id,
+          question_consolidator_llm_id: llm_model.id,
           forced_tool_count: 2,
         )
       persona2.create_user!
@@ -60,7 +66,8 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
       expect(serializer_persona2["allow_chat_channel_mentions"]).to eq(true)
       expect(serializer_persona2["allow_chat_direct_messages"]).to eq(true)
 
-      expect(serializer_persona2["default_llm"]).to eq("anthropic:claude-2")
+      expect(serializer_persona2["default_llm_id"]).to eq(llm_model.id)
+      expect(serializer_persona2["question_consolidator_llm_id"]).to eq(llm_model.id)
       expect(serializer_persona2["user_id"]).to eq(persona2.user_id)
       expect(serializer_persona2["user"]["id"]).to eq(persona2.user_id)
       expect(serializer_persona2["forced_tool_count"]).to eq(2)
@@ -178,7 +185,8 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
           allow_personal_messages: true,
           allow_chat_channel_mentions: true,
           allow_chat_direct_messages: true,
-          default_llm: "anthropic:claude-2",
+          default_llm_id: llm_model.id,
+          question_consolidator_llm_id: llm_model.id,
           forced_tool_count: 2,
         }
       end
@@ -190,18 +198,20 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
                headers: {
                  "CONTENT_TYPE" => "application/json",
                }
+
           expect(response).to be_successful
           persona_json = response.parsed_body["ai_persona"]
 
           expect(persona_json["name"]).to eq("superbot")
           expect(persona_json["top_p"]).to eq(0.1)
           expect(persona_json["temperature"]).to eq(0.5)
-          expect(persona_json["default_llm"]).to eq("anthropic:claude-2")
+          expect(persona_json["default_llm_id"]).to eq(llm_model.id)
           expect(persona_json["forced_tool_count"]).to eq(2)
           expect(persona_json["allow_topic_mentions"]).to eq(true)
           expect(persona_json["allow_personal_messages"]).to eq(true)
           expect(persona_json["allow_chat_channel_mentions"]).to eq(true)
           expect(persona_json["allow_chat_direct_messages"]).to eq(true)
+          expect(persona_json["question_consolidator_llm_id"]).to eq(llm_model.id)
 
           persona = AiPersona.find(persona_json["id"])
 
@@ -258,6 +268,8 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
               rag_chunk_tokens: "102",
               rag_chunk_overlap_tokens: "12",
               rag_conversation_chunks: "13",
+              rag_llm_model_id: llm_model.id,
+              question_consolidator_llm_id: llm_model.id,
             },
           }
 
@@ -267,6 +279,8 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
       expect(persona.rag_chunk_tokens).to eq(102)
       expect(persona.rag_chunk_overlap_tokens).to eq(12)
       expect(persona.rag_conversation_chunks).to eq(13)
+      expect(persona.rag_llm_model_id).to eq(llm_model.id)
+      expect(persona.question_consolidator_llm_id).to eq(llm_model.id)
     end
 
     it "supports updating vision params" do
@@ -424,7 +438,7 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
     end
 
     it "ensures question exists" do
-      ai_persona.update!(default_llm: "custom:#{llm.id}")
+      ai_persona.update!(default_llm_id: llm.id)
 
       post "/admin/plugins/discourse-ai/ai-personas/stream-reply.json",
            params: {
@@ -436,7 +450,7 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
     end
 
     it "ensure persona has a user specified" do
-      ai_persona.update!(default_llm: "custom:#{llm.id}")
+      ai_persona.update!(default_llm_id: llm.id)
 
       post "/admin/plugins/discourse-ai/ai-personas/stream-reply.json",
            params: {
@@ -498,7 +512,7 @@ RSpec.describe DiscourseAi::Admin::AiPersonasController do
       ai_persona.create_user!
       ai_persona.update!(
         allowed_group_ids: [Group::AUTO_GROUPS[:trust_level_0]],
-        default_llm: "custom:#{llm.id}",
+        default_llm_id: llm.id,
         allow_personal_messages: true,
         system_prompt: "you are a helpful bot",
       )
diff --git a/spec/requests/admin/ai_tools_controller_spec.rb b/spec/requests/admin/ai_tools_controller_spec.rb
index ef9d8737..fe5a9d4c 100644
--- a/spec/requests/admin/ai_tools_controller_spec.rb
+++ b/spec/requests/admin/ai_tools_controller_spec.rb
@@ -1,6 +1,7 @@
 # frozen_string_literal: true
 
 RSpec.describe DiscourseAi::Admin::AiToolsController do
+  fab!(:llm_model)
   fab!(:admin)
   fab!(:ai_tool) do
     AiTool.create!(
@@ -33,6 +34,7 @@ RSpec.describe DiscourseAi::Admin::AiToolsController do
       expect(response).to be_successful
       expect(response.parsed_body["ai_tools"].length).to eq(AiTool.count)
       expect(response.parsed_body["meta"]["presets"].length).to be > 0
+      expect(response.parsed_body["meta"]["llms"].length).to be > 0
     end
   end
 
diff --git a/spec/requests/admin/rag_document_fragments_controller_spec.rb b/spec/requests/admin/rag_document_fragments_controller_spec.rb
index c0e7d4df..7a906aaf 100644
--- a/spec/requests/admin/rag_document_fragments_controller_spec.rb
+++ b/spec/requests/admin/rag_document_fragments_controller_spec.rb
@@ -12,6 +12,8 @@ RSpec.describe DiscourseAi::Admin::RagDocumentFragmentsController do
     SiteSetting.ai_embeddings_enabled = true
   end
 
+  after { @cleanup_files&.each(&:unlink) }
+
   describe "GET #indexing_status_check" do
     it "works for AiPersona" do
       get "/admin/plugins/discourse-ai/rag-document-fragments/files/status.json?target_type=AiPersona&target_id=#{ai_persona.id}"
@@ -22,6 +24,15 @@ RSpec.describe DiscourseAi::Admin::RagDocumentFragmentsController do
   end
 
   describe "POST #upload_file" do
+    let :fake_pdf do
+      @cleanup_files ||= []
+      tempfile = Tempfile.new(%w[test .pdf])
+      tempfile.write("fake pdf")
+      tempfile.rewind
+      @cleanup_files << tempfile
+      tempfile
+    end
+
     it "works" do
       post "/admin/plugins/discourse-ai/rag-document-fragments/files/upload.json",
            params: {
@@ -33,5 +44,28 @@ RSpec.describe DiscourseAi::Admin::RagDocumentFragmentsController do
       upload = Upload.last
       expect(upload.original_filename).to end_with("spec.txt")
     end
+
+    it "rejects PDF files if site setting is not enabled" do
+      SiteSetting.ai_rag_pdf_images_enabled = false
+
+      post "/admin/plugins/discourse-ai/rag-document-fragments/files/upload.json",
+           params: {
+             file: Rack::Test::UploadedFile.new(fake_pdf),
+           }
+
+      expect(response.status).to eq(400)
+    end
+
+    it "allows PDF files if site setting is enabled" do
+      SiteSetting.ai_rag_pdf_images_enabled = true
+
+      post "/admin/plugins/discourse-ai/rag-document-fragments/files/upload.json",
+           params: {
+             file: Rack::Test::UploadedFile.new(fake_pdf),
+           }
+
+      upload = Upload.last
+      expect(upload.original_filename).to end_with(".pdf")
+    end
   end
 end
diff --git a/spec/system/ai_bot/ai_bot_helper_spec.rb b/spec/system/ai_bot/ai_bot_helper_spec.rb
index 97186a05..d7762d09 100644
--- a/spec/system/ai_bot/ai_bot_helper_spec.rb
+++ b/spec/system/ai_bot/ai_bot_helper_spec.rb
@@ -39,7 +39,7 @@ RSpec.describe "AI chat channel summarization", type: :system, js: true do
 
     # lets disable bots but still allow 1 persona
     allowed_persona.create_user!
-    allowed_persona.update!(default_llm: "custom:#{gpt_4.id}")
+    allowed_persona.update!(default_llm_id: gpt_4.id)
 
     gpt_4.update!(enabled_chat_bot: false)
     gpt_3_5_turbo.update!(enabled_chat_bot: false)
diff --git a/test/javascripts/unit/models/ai-persona-test.js b/test/javascripts/unit/models/ai-persona-test.js
index c8178f19..0ed6fe66 100644
--- a/test/javascripts/unit/models/ai-persona-test.js
+++ b/test/javascripts/unit/models/ai-persona-test.js
@@ -37,7 +37,7 @@ module("Discourse AI | Unit | Model | ai-persona", function () {
       description: "Description",
       top_p: 0.8,
       temperature: 0.7,
-      default_llm: "Default LLM",
+      default_llm_id: 1,
       force_default_llm: false,
       user: null,
       user_id: null,
@@ -48,7 +48,8 @@ module("Discourse AI | Unit | Model | ai-persona", function () {
       rag_chunk_tokens: 374,
       rag_chunk_overlap_tokens: 10,
       rag_conversation_chunks: 10,
-      question_consolidator_llm: "Question Consolidator LLM",
+      rag_llm_model_id: 1,
+      question_consolidator_llm_id: 2,
       allow_chat: false,
       tool_details: true,
       forced_tool_count: -1,
@@ -85,7 +86,7 @@ module("Discourse AI | Unit | Model | ai-persona", function () {
       temperature: 0.7,
       user: null,
       user_id: null,
-      default_llm: "Default LLM",
+      default_llm_id: 1,
       max_context_posts: 5,
       vision_enabled: true,
       vision_max_pixels: 100,
@@ -93,7 +94,7 @@ module("Discourse AI | Unit | Model | ai-persona", function () {
       rag_chunk_tokens: 374,
       rag_chunk_overlap_tokens: 10,
       rag_conversation_chunks: 10,
-      question_consolidator_llm: "Question Consolidator LLM",
+      question_consolidator_llm_id: 2,
       allow_chat: false,
       tool_details: true,
       forced_tool_count: -1,
@@ -102,6 +103,7 @@ module("Discourse AI | Unit | Model | ai-persona", function () {
       allow_chat_channel_mentions: true,
       allow_chat_direct_messages: true,
       force_default_llm: false,
+      rag_llm_model_id: 1,
     };
 
     const aiPersona = AiPersona.create({ ...properties });