Ai customization settings (#7069)

# Description of Changes
AI settings customisation in settings menu, as part of this also tested
and fixed ollama and other 3rd party AI integrations

- Adds an admin AI settings UI for customizing AI behaviour, including
per-provider model and API-key configuration
- Backend pushes AI config changes to the Python engine at runtime via a
config-push bridge, so changes apply without a restart
- Config-push is gated off in SaaS; engine now drains background tasks
on shutdown instead of cancelling them
---

## Checklist

### General

- [ ] I have read the [Contribution
Guidelines](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/CONTRIBUTING.md)
- [ ] I have read the [Stirling-PDF Developer
Guide](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/DeveloperGuide.md)
(if applicable)
- [ ] I have read the [How to add new languages to
Stirling-PDF](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/devGuide/HowToAddNewLanguage.md)
(if applicable)
- [ ] I have performed a self-review of my own code
- [ ] My changes generate no new warnings

### Documentation

- [ ] I have updated relevant docs on [Stirling-PDF's doc
repo](https://github.com/Stirling-Tools/Stirling-Tools.github.io/blob/main/docs/)
(if functionality has heavily changed)
- [ ] I have read the section [Add New Translation
Tags](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/devGuide/HowToAddNewLanguage.md#add-new-translation-tags)
(for new translation tags only)

### Translations (if applicable)

- [ ] I ran
[`scripts/counter_translation.py`](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/docs/counter_translation.md)

### UI Changes (if applicable)

- [ ] Screenshots or videos demonstrating the UI changes are attached
(e.g., as comments or direct attachments in the PR)

### Testing (if applicable)

- [ ] I have run `task check` to verify linters, typechecks, and tests
pass
- [ ] I have tested my changes locally. Refer to the [Testing
Guide](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/DeveloperGuide.md#7-testing)
for more details.

app/core/src/main/java/stirling/software/SPDF/controller/api/misc/ConfigController.java

@@ -336,7 +336,19 @@ public class ConfigController {
             configData.put("premiumEnabled", applicationProperties.getPremium().isEnabled());
 
             // AI Engine settings
-            configData.put("aiEngineEnabled", applicationProperties.getAiEngine().isEnabled());
+            ApplicationProperties.AiEngine aiEngineConfig = applicationProperties.getAiEngine();
+            configData.put("aiEngineEnabled", aiEngineConfig.isEnabled());
+            // Per-capability flags let the UI hide individual AI tools an admin has turned off.
+            ApplicationProperties.AiEngine.Features aiFeatures = aiEngineConfig.getFeatures();
+            configData.put(
+                    "aiFeatures",
+                    Map.ofEntries(
+                            Map.entry("chat", aiFeatures.isChat()),
+                            Map.entry("documentQuestions", aiFeatures.isDocumentQuestions()),
+                            Map.entry("createPdf", aiFeatures.isCreatePdf()),
+                            Map.entry("mathAuditor", aiFeatures.isMathAuditor()),
+                            Map.entry("pdfComment", aiFeatures.isPdfComment()),
+                            Map.entry("classify", aiFeatures.isClassify())));
 
             // Timestamp TSA settings — single source of truth for presets + admin URLs
             ApplicationProperties.Security.Timestamp tsConfig =

app/core/src/main/resources/settings.yml.template

@@ -366,6 +366,35 @@ aiEngine:
   enabled: false # Set to 'true' to enable the AI engine integration
   url: http://localhost:5001 # URL of the Python AI engine
   timeoutSeconds: 120 # Timeout in seconds for AI engine requests
+  longRunningTimeoutSeconds: 600 # Timeout (seconds) for heavy operations like RAG ingestion of large documents
+  streamTimeoutSeconds: 1800 # SSE stream timeout (seconds) for long-running orchestrator runs
+  pushConfigToEngine: true # Push admin AI config to the engine on startup + save; false = engine stays fully env-controlled
+  models:
+    provider: anthropic # Model provider: 'anthropic', 'openai', 'ollama', or 'custom' (OpenAI-compatible)
+    smartModel: claude-haiku-4-5 # High-quality tier model name (no provider prefix)
+    fastModel: claude-haiku-4-5 # Cheap/fast tier model name (no provider prefix)
+    smartMaxTokens: 8192 # Max output tokens for the smart tier
+    fastMaxTokens: 2048 # Max output tokens for the fast tier
+    apiKey: "" # API key for the selected provider (secret). Empty = engine uses its native env credentials (e.g. ANTHROPIC_API_KEY)
+    baseUrl: "" # OpenAI-compatible base URL for 'ollama'/'custom' providers (e.g. http://ollama:11434/v1). Ignored for anthropic/openai
+  rag:
+    embeddingProvider: voyageai # Embedding provider: 'voyageai', 'openai', 'ollama', or 'custom' (OpenAI-compatible)
+    embeddingModel: voyage-4 # Embedding model name (no provider prefix)
+    embeddingApiKey: "" # Secret API key for the embedding provider. Empty = engine uses its native env credentials (e.g. VOYAGE_API_KEY)
+    embeddingBaseUrl: "" # OpenAI-compatible base URL for 'ollama'/'custom' embedding providers (e.g. http://ollama:11434/v1). Ignored for voyageai/openai
+    topK: 20 # Number of chunks retrieval returns per search
+    maxSearches: 5 # Per-run cap on knowledge-search tool calls before the agent must answer
+  limits:
+    maxPages: 200 # Upper bound on PDF pages the engine will process per request
+    maxCharacters: 200000 # Upper bound on characters of extracted text per request
+    modelMaxConcurrency: 32 # Process-wide cap on concurrent model API calls (engine restart to apply)
+  features: # Per-capability switches; turn an individual AI tool off without disabling the whole engine
+    chat: true # Assistant chat
+    documentQuestions: true # Ask-questions-about-a-PDF
+    createPdf: true # Generate a PDF from a natural-language spec
+    mathAuditor: true # Numerical/formula contradiction auditing
+    pdfComment: true # AI-authored PDF comments/annotations
+    classify: true # Automatic document classification/labelling
 
 policies:
   # Folder automations can read from and write to the directories you allow here, so treat this as a
@@ -385,6 +414,8 @@ policies:
 mcp:
   enabled: false # Master switch. 'false' (default) means no /mcp endpoint, no metadata, no beans wired.
   scopesEnabled: true # Enforce mcp.tools.read / mcp.tools.write scopes derived from operation category
+  maxRequestBytes: 10485760 # Max size (bytes) of an incoming MCP tool request payload (default 10 MB)
+  maxInlineResponseBytes: 10485760 # Max size (bytes) of an MCP tool response returned inline before it is rejected (default 10 MB)
   allowedOperations: [] # Tool allow-list (operation ids, e.g. ['compress-pdf']). Empty = all. When set, ONLY these are exposed over MCP.
   blockedOperations: [] # Tool deny-list (operation ids). Always removed from MCP even if otherwise allowed.
   auth: