fix: quarantine dead brand commands + implement Memory LLM summary

Brand commands (~290 lines):
- Code was trapped inside `if reply_to_message:` block (unreachable)
- Moved to feature flag: ENABLE_BRAND_COMMANDS=true to activate
- Zero re-indentation: 8sp code naturally fits as feature flag body
- Helper functions (_brand_*, _artifact_*) unchanged

Memory LLM Summary:
- Replace placeholder with real DeepSeek API integration
- Structured output: summary, goals, decisions, open_questions, next_steps, key_facts
- Graceful fallback if API key not set or call fails
- Added MEMORY_DEEPSEEK_API_KEY config
- Ukrainian output language

Deployed and verified on NODE1.

Co-authored-by: Cursor <cursoragent@cursor.com>

gateway-bot/http_api.py

@@ -1365,6 +1365,13 @@ async def handle_telegram_webhook(
                     f"(user {username} replied to bot msg {reply_to_message.get('message_id', '?')})"
                 )
 
+    # === BRAND COMMANDS (experimental, disabled by default) ===
+    # ~290 lines of brand/presentation command handlers.
+    # This code was NEVER active in production (was trapped inside wrong indent block).
+    # Quarantined behind feature flag. Enable with: ENABLE_BRAND_COMMANDS=true
+    # See: /бренд, /бренд_інтейк, /бренд_тема, /презентація, /job_статус
+    _brand_commands_enabled = os.environ.get("ENABLE_BRAND_COMMANDS", "").lower() in ("1", "true", "yes")
+    if _brand_commands_enabled:
         text = update.message.get("text", "")
 
         # Simple brand commands (Ukrainian)

services/memory-service/app/config.py

@@ -34,6 +34,12 @@ class Settings(BaseSettings):
     cohere_model: str = "embed-multilingual-v3.0"  # 1024 dimensions
     embedding_dimensions: int = 1024
     
+    # DeepSeek LLM (for summary generation)
+    deepseek_api_key: str = ""  # Must be set via MEMORY_DEEPSEEK_API_KEY env var
+    deepseek_api_url: str = "https://api.deepseek.com/v1/chat/completions"
+    deepseek_model: str = "deepseek-chat"
+    summary_max_tokens: int = 800
+    
     # Memory settings
     short_term_window_messages: int = 20
     short_term_window_minutes: int = 60

services/memory-service/app/main.py

@@ -7,10 +7,11 @@ DAARION Memory Service - FastAPI Application
 - Long-term: memory items (персональна/проектна)
 """
 from contextlib import asynccontextmanager
-from typing import List, Optional
+from typing import List, Optional, Dict, Any
 from fastapi import Depends, BackgroundTasks
 from uuid import UUID
 import structlog
+import httpx
 from fastapi import FastAPI, HTTPException, Query
 from fastapi.middleware.cors import CORSMiddleware
 
@@ -30,6 +31,110 @@ logger = structlog.get_logger()
 settings = get_settings()
 
 
+SUMMARY_SYSTEM_PROMPT = """You are a conversation summarizer for DAARION AI agents.
+Given a list of conversation events, produce a structured summary in Ukrainian.
+
+Output format (JSON):
+{
+  "summary": "Стислий опис розмови (2-5 речень)",
+  "goals": ["Ціль 1", "Ціль 2"],
+  "decisions": ["Рішення 1", "Рішення 2"],
+  "open_questions": ["Питання 1"],
+  "next_steps": ["Крок 1", "Крок 2"],
+  "key_facts": ["Факт 1", "Факт 2"]
+}
+
+Rules:
+- Write in Ukrainian
+- Be concise but preserve key information
+- Focus on actionable items and decisions
+- If no items for a category, use empty array []
+- Return ONLY valid JSON, no markdown
+"""
+
+
+async def _llm_generate_summary(events: List[Dict[str, Any]]) -> Dict[str, Any]:
+    """
+    Call DeepSeek API to generate structured thread summary.
+    Falls back to placeholder if API key is not configured or call fails.
+    """
+    if not settings.deepseek_api_key:
+        logger.warning("deepseek_api_key_not_set", hint="Set MEMORY_DEEPSEEK_API_KEY to enable LLM summaries")
+        return {
+            "summary": f"Summary of {len(events)} events. [LLM not configured — set MEMORY_DEEPSEEK_API_KEY]",
+            "goals": [],
+            "decisions": [],
+            "open_questions": [],
+            "next_steps": [],
+            "key_facts": []
+        }
+
+    # Format events for LLM
+    events_text = []
+    for ev in events[-50:]:  # Limit to last 50 events to stay within context
+        role = ev.get("role", "unknown")
+        content = ev.get("content", "")[:500]  # Truncate long messages
+        events_text.append(f"[{role}]: {content}")
+    conversation = "\n".join(events_text)
+
+    try:
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            resp = await client.post(
+                settings.deepseek_api_url,
+                headers={
+                    "Authorization": f"Bearer {settings.deepseek_api_key}",
+                    "Content-Type": "application/json",
+                },
+                json={
+                    "model": settings.deepseek_model,
+                    "messages": [
+                        {"role": "system", "content": SUMMARY_SYSTEM_PROMPT},
+                        {"role": "user", "content": f"Summarize this conversation:\n\n{conversation}"},
+                    ],
+                    "max_tokens": settings.summary_max_tokens,
+                    "temperature": 0.3,
+                },
+            )
+            resp.raise_for_status()
+            data = resp.json()
+            content = data["choices"][0]["message"]["content"]
+
+            # Parse JSON response
+            import json
+            try:
+                parsed = json.loads(content)
+                return {
+                    "summary": parsed.get("summary", content),
+                    "goals": parsed.get("goals", []),
+                    "decisions": parsed.get("decisions", []),
+                    "open_questions": parsed.get("open_questions", []),
+                    "next_steps": parsed.get("next_steps", []),
+                    "key_facts": parsed.get("key_facts", []),
+                }
+            except json.JSONDecodeError:
+                # LLM returned non-JSON — use raw text as summary
+                logger.warning("llm_summary_not_json", content_preview=content[:100])
+                return {
+                    "summary": content,
+                    "goals": [],
+                    "decisions": [],
+                    "open_questions": [],
+                    "next_steps": [],
+                    "key_facts": [],
+                }
+
+    except Exception as e:
+        logger.error("llm_summary_failed", error=str(e))
+        return {
+            "summary": f"Summary of {len(events)} events. [LLM call failed: {str(e)[:100]}]",
+            "goals": [],
+            "decisions": [],
+            "open_questions": [],
+            "next_steps": [],
+            "key_facts": [],
+        }
+
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     """Startup and shutdown events"""
@@ -376,15 +481,15 @@ async def create_summary(thread_id: UUID, request: SummaryRequest):
     # Get events to summarize
     events = await db.get_events_for_summary(thread_id)
     
-    # TODO: Call LLM to generate summary
+    # Generate structured summary via DeepSeek LLM
-    # For now, create a placeholder
+    llm_result = await _llm_generate_summary(events)
-    summary_text = f"Summary of {len(events)} events. [Implement LLM summarization]"
+    summary_text = llm_result["summary"]
     state = {
-        "goals": [],
+        "goals": llm_result["goals"],
-        "decisions": [],
+        "decisions": llm_result["decisions"],
-        "open_questions": [],
+        "open_questions": llm_result["open_questions"],
-        "next_steps": [],
+        "next_steps": llm_result["next_steps"],
-        "key_facts": []
+        "key_facts": llm_result["key_facts"]
     }
     
     # Create summary