docs: add Docker Hub publish workflow and git mirror remotes

- Layer 2 semantic search now queries both curated and raw types,
  closing the blind spot for turns past the 50-turn window pre-curation
- Layer 3 skips turns already returned by Layer 2 to avoid duplicate
  context and wasted token budget
- get_recent_turns uses Qdrant OrderBy for server-side timestamp sort
  with payload index; fallback to client-side sort if unavailable
- Bump version to 2.0.4

CLAUDE.md

@@ -1,128 +1,187 @@
-# Vera-AI Project
+# CLAUDE.md
 
-**Persistent Memory Proxy for Ollama**
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
 
-> **Status:** Built and running on deb8. Goal: Validate and improve.
+## Infrastructure
 
-Vera-AI sits between AI clients and Ollama, storing conversations in Qdrant and retrieving context semantically — giving AI **true memory**.
+| Role | Host | Access |
+|------|------|--------|
+| Source (deb9) | 10.0.0.48 | `ssh deb9` — `/home/n8n/vera-ai/` |
+| Production (deb8) | 10.0.0.46 | `ssh deb8` — runs vera-ai in Docker |
+| Gitea | 10.0.0.61:3000 | `SpeedyFoxAi/vera-ai-v2`, HTTPS only (SSH disabled) |
+
+User `n8n` on deb8/deb9. SSH key `~/.ssh/vera-ai`. Gitea credentials in `~/.netrc`.
+
+## Git Workflow
+
+Three locations — all point to `origin` on Gitea:
+
+```
+local (/home/adm1n/claude/vera-ai)  ←→  Gitea (10.0.0.61:3000)  ←→  deb9 (/home/n8n/vera-ai)
+        ↓                                                                   ↓
+  github/gitlab                                                  deb8 (scp files + docker build)
+  (mirrors)
+```
+
+```bash
+# Edit on deb9, commit, push
+ssh deb9
+cd /home/n8n/vera-ai
+git pull origin main              # sync first
+git add -p && git commit -m "..."
+git push origin main
+
+# Pull to local working copy
+cd /home/adm1n/claude/vera-ai
+git pull origin main
+
+# Deploy to production (deb8 has no git repo — scp files then build)
+scp app/*.py n8n@10.0.0.46:/home/n8n/vera-ai/app/
+ssh deb8 'cd /home/n8n/vera-ai && docker compose build && docker compose up -d'
+```
+
+## Publishing (Docker Hub + Git Mirrors)
+
+Image: `mdkrushr/vera-ai` on Docker Hub. Build and push from deb8:
+
+```bash
+ssh deb8
+cd /home/n8n/vera-ai
+docker build -t mdkrushr/vera-ai:2.0.4 -t mdkrushr/vera-ai:latest .
+docker push mdkrushr/vera-ai:2.0.4
+docker push mdkrushr/vera-ai:latest
+```
+
+The local repo has two mirror remotes for public distribution. After committing and pushing to `origin` (Gitea), mirror with:
+
+```bash
+git push github main --tags
+git push gitlab main --tags
+```
+
+| Remote | URL |
+|--------|-----|
+| `origin` | `10.0.0.61:3000/SpeedyFoxAi/vera-ai-v2` (Gitea, primary) |
+| `github` | `github.com/speedyfoxai/vera-ai` |
+| `gitlab` | `gitlab.com/mdkrush/vera-ai` |
+
+## Build & Run (deb8, production)
+
+```bash
+ssh deb8
+cd /home/n8n/vera-ai
+docker compose build
+docker compose up -d
+docker logs vera-ai --tail 30
+curl http://localhost:11434/                   # health check
+curl -X POST http://localhost:11434/curator/run  # trigger curation
+```
+
+## Tests (deb9, source)
+
+```bash
+ssh deb9
+cd /home/n8n/vera-ai
+python3 -m pytest tests/                                          # all tests
+python3 -m pytest tests/test_utils.py                             # single file
+python3 -m pytest tests/test_utils.py::TestParseCuratedTurn::test_single_turn  # single test
+python3 -m pytest tests/ --cov=app --cov-report=term-missing      # with coverage
+```
+
+Tests are unit-only — no live Qdrant/Ollama required. `pytest.ini` sets `asyncio_mode=auto`. Shared fixtures with production-realistic data in `tests/conftest.py`.
+
+Test files and what they cover:
+
+| File | Covers |
+|------|--------|
+| `tests/test_utils.py` | Token counting, truncation, memory filtering/merging, `parse_curated_turn`, `load_system_prompt`, `build_augmented_messages` |
+| `tests/test_config.py` | Config defaults, TOML loading, `CloudConfig`, env var overrides |
+| `tests/test_curator.py` | JSON parsing, `_is_recent`, `_format_raw_turns`, `_format_existing_memories`, `_call_llm`, `_append_rule_to_file`, `load_curator_prompt`, full `run()` scenarios |
+| `tests/test_proxy_handler.py` | `clean_message_content`, `handle_chat_non_streaming`, `debug_log`, `forward_to_ollama` |
+| `tests/test_integration.py` | FastAPI health check, `/api/tags` (with cloud models), `/api/chat` round-trips (streaming + non-streaming), curator trigger, proxy passthrough |
+| `tests/test_qdrant_service.py` | `_ensure_collection`, `get_embedding`, `store_turn`, `store_qa_turn`, `semantic_search`, `get_recent_turns`, `delete_points`, `close` |
 
 ## Architecture
 
 ```
-Client → Vera-AI (port 11434) → Ollama
+Client → Vera-AI :11434 → Ollama :11434
-              ↓
+               ↓↑
-           Qdrant (vector DB)
+          Qdrant :6333
-              ↓
-           Memory Storage
 ```
 
-## Key Components
+Vera-AI is a FastAPI proxy. Every `/api/chat` request is intercepted, augmented with memory context, forwarded to Ollama, and the response Q&A is stored back in Qdrant.
 
-| File | Purpose |
+### 4-Layer Context System (`app/utils.py:build_augmented_messages`)
-|------|---------|
-| `app/main.py` | FastAPI application entry point |
-| `app/proxy_handler.py` | Chat request handling |
-| `app/qdrant_service.py` | Vector DB operations |
-| `app/curator.py` | Memory curation (daily/monthly) |
-| `app/config.py` | Configuration loader |
-| `config/config.toml` | Main configuration file |
 
-## 4-Layer Context System
+Each chat request builds an augmented message list in this order:
 
-1. **System Prompt** — From `prompts/systemprompt.md`
+1. **System** — caller's system prompt passed through; `prompts/systemprompt.md` appended if non-empty (if empty, caller's prompt passes through unchanged; if no caller system prompt, vera's prompt used alone)
-2. **Semantic Memory** — Curated Q&A from Qdrant (relevance search)
+2. **Semantic** — curated AND raw Q&A pairs from Qdrant matching the query (score ≥ `semantic_score_threshold`, up to `semantic_token_budget` tokens). Searches both types to avoid a blind spot where raw turns fall off the recent window before curation runs.
-3. **Recent Context** — Last N conversation turns
+3. **Recent context** — last 50 turns from Qdrant (server-sorted by timestamp via payload index), oldest first, up to `context_token_budget` tokens. Deduplicates against Layer 2 results to avoid wasting token budget.
-4. **Current Messages** — User's current request
+4. **Current** — the incoming messages (non-system) passed through unchanged
+
+The system prompt is **never truncated**. Semantic and context layers are budget-limited and drop excess entries silently.
+
+### Memory Types in Qdrant
+
+| Type | When created | Retention |
+|------|-------------|-----------|
+| `raw` | After each chat turn | Until curation runs |
+| `curated` | After curator processes `raw` | Permanent |
+
+Payload format: `{type, text, timestamp, role, content}`. Curated entries use `role="curated"` with text formatted as `User: ...\nAssistant: ...\nTimestamp: ...`, which `parse_curated_turn()` deserializes back into proper message role pairs at retrieval time.
+
+### Curator (`app/curator.py`)
+
+Scheduled via APScheduler at `config.run_time` (default 02:00). Automatically detects day 01 of month for monthly mode (processes ALL raw) vs. daily mode (last 24h only). Sends raw memories to `curator_model` LLM with `prompts/curator_prompt.md`, expects JSON response:
+
+```json
+{
+  "new_curated_turns": [{"content": "User: ...\nAssistant: ..."}],
+  "permanent_rules": [{"rule": "...", "target_file": "systemprompt.md"}],
+  "deletions": ["uuid1", "uuid2"],
+  "summary": "..."
+}
+```
+
+`permanent_rules` are appended to the named file in `prompts/`. After curation, all processed raw entries are deleted.
+
+### Cloud Model Routing
+
+Optional `[cloud]` section in `config.toml` routes specific model names to an OpenRouter-compatible API instead of Ollama. Cloud models are injected into `/api/tags` so clients see them alongside local models.
+
+```toml
+[cloud]
+enabled = true
+api_base = "https://openrouter.ai/api/v1"
+api_key_env = "OPENROUTER_API_KEY"
+[cloud.models]
+"gpt-oss:120b" = "openai/gpt-4o"
+```
+
+### Key Implementation Details
+
+- **Config loading** uses stdlib `tomllib` (read-only, Python 3.11+). No third-party TOML dependency.
+- **QdrantService singleton** lives in `app/singleton.py`. All modules import from there — `app/utils.py` re-exports via `from .singleton import get_qdrant_service`.
+- **Datetime handling** uses `datetime.now(timezone.utc)` throughout. No `utcnow()` calls. Stored timestamps are naive UTC with "Z" suffix; comparison code strips tzinfo for naive-vs-naive matching.
+- **Debug logging** in `proxy_handler.py` uses `portalocker` for file locking under concurrent requests. Controlled by `config.debug`.
 
 ## Configuration
 
-Key settings in `config/config.toml`:
+All settings in `config/config.toml`. Key tuning knobs:
 
-```toml
+- `semantic_token_budget` / `context_token_budget` — controls how much memory gets injected
-[general]
+- `semantic_score_threshold` — lower = more (but less relevant) memories returned
-ollama_host = "http://10.0.0.10:11434"
+- `curator_model` — model used for daily curation (needs strong reasoning)
-qdrant_host = "http://10.0.0.22:6333"
+- `debug = true` — enables per-request JSON logs written to `logs/debug_YYYY-MM-DD.log`
-qdrant_collection = "memories"
-embedding_model = "snowflake-arctic-embed2"
 
-[layers]
+Environment variable overrides: `VERA_CONFIG_DIR`, `VERA_PROMPTS_DIR`, `VERA_LOG_DIR`.
-semantic_token_budget = 25000
-context_token_budget = 22000
-semantic_search_turns = 2
-semantic_score_threshold = 0.6
 
-[curator]
+## Related Services
-run_time = "02:00"  # Daily curation time
-curator_model = "gpt-oss:120b"
-```
-
-## Environment Variables
-
-| Variable | Default | Description |
-|----------|---------|-------------|
-| `APP_UID` | `999` | Container user ID |
-| `APP_GID` | `999` | Container group ID |
-| `TZ` | `UTC` | Timezone |
-| `VERA_DEBUG` | `false` | Enable debug logging |
-
-## Running
-
-```bash
-# Build and start
-docker compose build
-docker compose up -d
-
-# Check status
-docker ps
-docker logs VeraAI --tail 20
-
-# Health check
-curl http://localhost:11434/
-```
-
-## API Endpoints
-
-| Endpoint | Method | Description |
-|----------|--------|-------------|
-| `/` | GET | Health check |
-| `/api/chat` | POST | Chat completion (with memory) |
-| `/api/tags` | GET | List models |
-| `/api/generate` | POST | Generate completion |
-| `/curator/run` | POST | Trigger curation manually |
-
-## Development Workflow
-
-This project is synced with **deb9** (10.0.0.48). To sync changes:
-
-```bash
-# Pull from deb9
-sshpass -p 'passw0rd' scp -r -o StrictHostKeyChecking=no n8n@10.0.0.48:/home/n8n/vera-ai/* /home/n8n/vera-ai/
-
-# Push to deb9 (after local changes)
-sshpass -p 'passw0rd' scp -r -o StrictHostKeyChecking=no /home/n8n/vera-ai/* n8n@10.0.0.48:/home/n8n/vera-ai/
-```
-
-## Memory System
-
-- **raw** memories — Unprocessed conversation turns (until curation)
-- **curated** memories — Cleaned Q&A pairs (permanent)
-- **test** memories — Test entries (can be ignored)
-
-Curation runs daily at 02:00 and monthly on the 1st at 03:00.
-
-## Related Infrastructure
 
 | Service | Host | Port |
 |---------|------|------|
-| Qdrant | 10.0.0.22 | 6333 |
 | Ollama | 10.0.0.10 | 11434 |
-| deb9 | 10.0.0.48 | Source project (SSH) |
+| Qdrant | 10.0.0.22 | 6333 |
-| deb8 | 10.0.0.46 | Docker runtime |
 
-## Qdrant Collections
+Qdrant collections: `memories` (default), `vera_memories` (alternative), `python_kb` (reference patterns).
-
-| Collection | Purpose |
-|------------|---------|
-| `python_kb` | Python code patterns reference for this project |
-| `memories` | Conversation memory storage (default) |
-| `vera_memories` | Alternative memory collection |

app/config.py

@@ -1,5 +1,5 @@
 # app/config.py
-import toml
+import tomllib
 import os
 from pathlib import Path
 from dataclasses import dataclass, field
@@ -83,8 +83,8 @@ class Config:
         config = cls()
         
         if config_path.exists():
-            with open(config_path, "r") as f:
+            with open(config_path, "rb") as f:
-                data = toml.load(f)
+                data = tomllib.load(f)
             
             if "general" in data:
                 config.ollama_host = data["general"].get("ollama_host", config.ollama_host)
@@ -112,7 +112,14 @@ class Config:
                     api_key_env=cloud_data.get("api_key_env", "OPENROUTER_API_KEY"),
                     models=cloud_data.get("models", {})
                 )
-        
+
+        if config.cloud.enabled and not config.cloud.api_key:
+            import logging
+            logging.getLogger(__name__).warning(
+                "Cloud is enabled but API key env var '%s' is not set",
+                config.cloud.api_key_env
+            )
+
         return config
 
 config = Config.load()

app/curator.py

@@ -6,7 +6,7 @@ The prompt determines behavior based on current date.
 """
 import logging
 import os
-from datetime import datetime, timedelta
+from datetime import datetime, timedelta, timezone
 from typing import List, Dict, Any, Optional
 from pathlib import Path
 import httpx
@@ -49,7 +49,7 @@ class Curator:
         Otherwise runs daily mode (processes recent 24h only).
         The prompt determines behavior based on current date.
         """
-        current_date = datetime.utcnow()
+        current_date = datetime.now(timezone.utc)
         is_monthly = current_date.day == 1
         mode = "MONTHLY" if is_monthly else "DAILY"
         
@@ -169,7 +169,7 @@ Remember: Respond with ONLY valid JSON. No markdown, no explanations, just the J
             return True
         try:
             mem_time = datetime.fromisoformat(timestamp.replace("Z", "+00:00"))
-            cutoff = datetime.utcnow() - timedelta(hours=hours)
+            cutoff = datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=hours)
             return mem_time.replace(tzinfo=None) > cutoff
         except (ValueError, TypeError):
             logger.debug(f"Could not parse timestamp: {timestamp}")
@@ -212,7 +212,7 @@ Remember: Respond with ONLY valid JSON. No markdown, no explanations, just the J
                 result = response.json()
                 return result.get("response", "")
         except Exception as e:
-            logger.error(f"Failed to call LLM: {e}")
+            logger.error(f"LLM call failed: {e}", exc_info=True)
             return ""
 
     def _parse_json_response(self, response: str) -> Optional[Dict]:
@@ -223,6 +223,7 @@ Remember: Respond with ONLY valid JSON. No markdown, no explanations, just the J
         try:
             return json.loads(response)
         except json.JSONDecodeError:
+            logger.debug("Direct JSON parse failed, trying brace extraction")
             pass
 
         try:

app/main.py

@@ -4,7 +4,7 @@ from fastapi.responses import StreamingResponse, JSONResponse
 from contextlib import asynccontextmanager
 import httpx
 import logging
-from datetime import datetime
+from datetime import datetime, timezone
 
 from .config import config
 from .singleton import get_qdrant_service
@@ -68,7 +68,7 @@ async def lifespan(app: FastAPI):
     await qdrant_service.close()
 
 
-app = FastAPI(title="Vera-AI", version="2.0.0", lifespan=lifespan)
+app = FastAPI(title="Vera-AI", version="2.0.4", lifespan=lifespan)
 
 
 @app.get("/")
@@ -96,7 +96,7 @@ async def api_tags():
         for name in config.cloud.models.keys():
             data["models"].append({
                 "name": name,
-                "modified_at": "2026-03-25T00:00:00Z",
+                "modified_at": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
                 "size": 0,
                 "digest": "cloud",
                 "details": {"family": "cloud"}

app/proxy_handler.py

@@ -6,6 +6,7 @@ import json
 import re
 import logging
 import os
+import portalocker
 from pathlib import Path
 from .config import config
 from .singleton import get_qdrant_service
@@ -48,17 +49,17 @@ def debug_log(category: str, message: str, data: dict = None):
     if not config.debug:
         return
     
-    from datetime import datetime
+    from datetime import datetime, timezone
     
     # Create logs directory
     log_dir = DEBUG_LOG_DIR
     log_dir.mkdir(parents=True, exist_ok=True)
     
-    today = datetime.utcnow().strftime("%Y-%m-%d")
+    today = datetime.now(timezone.utc).strftime("%Y-%m-%d")
     log_path = log_dir / f"debug_{today}.log"
     
     entry = {
-        "timestamp": datetime.utcnow().isoformat() + "Z",
+        "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
         "category": category,
         "message": message
     }
@@ -66,7 +67,9 @@ def debug_log(category: str, message: str, data: dict = None):
         entry["data"] = data
     
     with open(log_path, "a") as f:
+        portalocker.lock(f, portalocker.LOCK_EX)
         f.write(json.dumps(entry) + "\n")
+        portalocker.unlock(f)
 
 
 async def handle_chat_non_streaming(body: dict):

app/qdrant_service.py

@@ -1,8 +1,8 @@
 """Qdrant service for memory storage - ASYNC VERSION."""
 from qdrant_client import AsyncQdrantClient
-from qdrant_client.models import Distance, VectorParams, PointStruct, Filter, FieldCondition, MatchValue
+from qdrant_client.models import Distance, VectorParams, PointStruct, Filter, FieldCondition, MatchValue, PayloadSchemaType
 from typing import List, Dict, Any, Optional
-from datetime import datetime
+from datetime import datetime, timezone
 import uuid
 import logging
 import httpx
@@ -34,6 +34,15 @@ class QdrantService:
                 vectors_config=VectorParams(size=self.vector_size, distance=Distance.COSINE)
             )
             logger.info(f"Created collection {self.collection} with vector size {self.vector_size}")
+        # Ensure payload index on timestamp for ordered scroll
+        try:
+            await self.client.create_payload_index(
+                collection_name=self.collection,
+                field_name="timestamp",
+                field_schema=PayloadSchemaType.KEYWORD
+            )
+        except Exception:
+            pass  # Index may already exist
         self._collection_ensured = True
 
     async def get_embedding(self, text: str) -> List[float]:
@@ -54,7 +63,7 @@ class QdrantService:
         point_id = str(uuid.uuid4())
         embedding = await self.get_embedding(content)
         
-        timestamp = datetime.utcnow().isoformat() + "Z"
+        timestamp = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
         text = content
         if role == "user":
             text = f"User: {content}"
@@ -85,7 +94,7 @@ class QdrantService:
         """Store a complete Q&A turn as one document."""
         await self._ensure_collection()
         
-        timestamp = datetime.utcnow().isoformat() + "Z"
+        timestamp = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
         text = f"User: {user_question}\nAssistant: {assistant_answer}\nTimestamp: {timestamp}"
         
         point_id = str(uuid.uuid4())
@@ -105,20 +114,28 @@ class QdrantService:
         )
         return point_id
 
-    async def semantic_search(self, query: str, limit: int = 10, score_threshold: float = 0.6, entry_type: str = "curated") -> List[Dict]:
+    async def semantic_search(self, query: str, limit: int = 10, score_threshold: float = 0.6, entry_type: str = "curated", entry_types: Optional[List[str]] = None) -> List[Dict]:
-        """Semantic search for relevant turns, filtered by type."""
+        """Semantic search for relevant turns, filtered by type(s)."""
         await self._ensure_collection()
-        
+
         embedding = await self.get_embedding(query)
-        
+
+        if entry_types and len(entry_types) > 1:
+            type_filter = Filter(
+                should=[FieldCondition(key="type", match=MatchValue(value=t)) for t in entry_types]
+            )
+        else:
+            filter_type = entry_types[0] if entry_types else entry_type
+            type_filter = Filter(
+                must=[FieldCondition(key="type", match=MatchValue(value=filter_type))]
+            )
+
         results = await self.client.query_points(
             collection_name=self.collection,
             query=embedding,
             limit=limit,
             score_threshold=score_threshold,
-            query_filter=Filter(
+            query_filter=type_filter
-                must=[FieldCondition(key="type", match=MatchValue(value=entry_type))]
-            )
         )
         
         return [{"id": str(r.id), "score": r.score, "payload": r.payload} for r in results.points]
@@ -126,21 +143,29 @@ class QdrantService:
     async def get_recent_turns(self, limit: int = 20) -> List[Dict]:
         """Get recent turns from Qdrant (both raw and curated)."""
         await self._ensure_collection()
-        
+
-        points, _ = await self.client.scroll(
+        try:
-            collection_name=self.collection,
+            from qdrant_client.models import OrderBy
-            limit=limit * 2,
+            points, _ = await self.client.scroll(
-            with_payload=True
+                collection_name=self.collection,
-        )
+                limit=limit,
-        
+                with_payload=True,
-        # Sort by timestamp descending
+                order_by=OrderBy(key="timestamp", direction="desc")
-        sorted_points = sorted(
+            )
-            points,
+        except Exception:
-            key=lambda p: p.payload.get("timestamp", ""),
+            # Fallback: fetch extra points and sort client-side
-            reverse=True
+            points, _ = await self.client.scroll(
-        )
+                collection_name=self.collection,
-        
+                limit=limit * 5,
-        return [{"id": str(p.id), "payload": p.payload} for p in sorted_points[:limit]]
+                with_payload=True
+            )
+            points = sorted(
+                points,
+                key=lambda p: p.payload.get("timestamp", ""),
+                reverse=True
+            )[:limit]
+
+        return [{"id": str(p.id), "payload": p.payload} for p in points]
 
     async def delete_points(self, point_ids: List[str]) -> None:
         """Delete points by ID."""

app/singleton.py

@@ -1,8 +1,9 @@
 """Global singleton instances for Vera-AI."""
+from typing import Optional
 from .qdrant_service import QdrantService
 from .config import config
 
-_qdrant_service: QdrantService = None
+_qdrant_service: Optional[QdrantService] = None
 
 
 def get_qdrant_service() -> QdrantService:

app/utils.py

@@ -1,9 +1,10 @@
 """Utility functions for vera-ai."""
 from .config import config
+from .singleton import get_qdrant_service
 import tiktoken
 import os
 from typing import List, Dict, Optional
-from datetime import datetime, timedelta
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
 
 # Use cl100k_base encoding (GPT-4 compatible)
@@ -13,24 +14,6 @@ ENCODING = tiktoken.get_encoding("cl100k_base")
 PROMPTS_DIR = Path(os.environ.get("VERA_PROMPTS_DIR", "/app/prompts"))
 STATIC_DIR = Path(os.environ.get("VERA_STATIC_DIR", "/app/static"))
 
-# Global qdrant_service instance for utils
-_qdrant_service = None
-
-def get_qdrant_service():
-    """Get or create the QdrantService singleton."""
-    global _qdrant_service
-    if _qdrant_service is None:
-        from .config import config
-        from .qdrant_service import QdrantService
-        _qdrant_service = QdrantService(
-            host=config.qdrant_host,
-            collection=config.qdrant_collection,
-            embedding_model=config.embedding_model,
-            vector_size=config.vector_size,
-            ollama_host=config.ollama_host
-        )
-    return _qdrant_service
-
 def count_tokens(text: str) -> int:
     """Count tokens in text."""
     if not text:
@@ -56,7 +39,7 @@ def truncate_by_tokens(text: str, max_tokens: int) -> str:
 
 def filter_memories_by_time(memories: List[Dict], hours: int = 24) -> List[Dict]:
     """Filter memories from the last N hours."""
-    cutoff = datetime.utcnow() - timedelta(hours=hours)
+    cutoff = datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=hours)
     filtered = []
     for mem in memories:
         ts = mem.get("timestamp")
@@ -64,7 +47,7 @@ def filter_memories_by_time(memories: List[Dict], hours: int = 24) -> List[Dict]
             try:
                 # Parse ISO timestamp
                 if isinstance(ts, str):
-                    mem_time = datetime.fromisoformat(ts.replace("Z", "+00:00").replace("+00:00", ""))
+                    mem_time = datetime.fromisoformat(ts.replace("Z", "")).replace(tzinfo=None)
                 else:
                     mem_time = ts
                 if mem_time > cutoff:
@@ -100,15 +83,6 @@ def merge_memories(memories: List[Dict]) -> Dict:
         "ids": ids
     }
 
-def calculate_token_budget(total_budget: int, system_ratio: float = 0.2, 
-                           semantic_ratio: float = 0.5, context_ratio: float = 0.3) -> Dict[int, int]:
-    """Calculate token budgets for each layer."""
-    return {
-        "system": int(total_budget * system_ratio),
-        "semantic": int(total_budget * semantic_ratio),
-        "context": int(total_budget * context_ratio)
-    }
-
 def load_system_prompt() -> str:
     """Load system prompt from prompts directory."""
     import logging
@@ -219,36 +193,45 @@ async def build_augmented_messages(incoming_messages: List[Dict]) -> List[Dict]:
     }
     
     # === LAYER 1: System Prompt ===
-    system_content = ""
+    # Caller's system message passes through; systemprompt.md appends if non-empty.
+    caller_system = ""
     for msg in incoming_messages:
         if msg.get("role") == "system":
-            system_content = msg.get("content", "")
+            caller_system = msg.get("content", "")
             break
-    
+
-    if system_prompt:
+    if caller_system and system_prompt:
-        system_content += "\n\n" + system_prompt
+        system_content = caller_system + "\n\n" + system_prompt
-    
+    elif caller_system:
+        system_content = caller_system
+    elif system_prompt:
+        system_content = system_prompt
+    else:
+        system_content = ""
+
     if system_content:
         messages.append({"role": "system", "content": system_content})
         logger.info(f"Layer 1 (system): {count_tokens(system_content)} tokens")
     
-    # === LAYER 2: Semantic (curated memories) ===
+    # === LAYER 2: Semantic (curated + raw memories) ===
     qdrant = get_qdrant_service()
     semantic_results = await qdrant.semantic_search(
         query=search_context if search_context else user_question,
         limit=20,
         score_threshold=config.semantic_score_threshold,
-        entry_type="curated"
+        entry_types=["curated", "raw"]
     )
     
     semantic_messages = []
     semantic_tokens_used = 0
-    
+    semantic_ids = set()
+
     for result in semantic_results:
+        semantic_ids.add(result.get("id"))
         payload = result.get("payload", {})
         text = payload.get("text", "")
         if text:
-            # Parse curated turn into proper user/assistant messages
+            # Parse curated/raw turn into proper user/assistant messages
             parsed = parse_curated_turn(text)
             for msg in parsed:
                 msg_tokens = count_tokens(msg.get("content", ""))
@@ -273,8 +256,10 @@ async def build_augmented_messages(incoming_messages: List[Dict]) -> List[Dict]:
     context_messages = []
     context_tokens_used = 0
     
-    # Process oldest first for chronological order
+    # Process oldest first for chronological order, skip duplicates from Layer 2
     for turn in reversed(recent_turns):
+        if turn.get("id") in semantic_ids:
+            continue
         payload = turn.get("payload", {})
         text = payload.get("text", "")
         entry_type = payload.get("type", "raw")

requirements.txt

@@ -1,11 +1,11 @@
-fastapi>=0.104.0
+fastapi==0.135.2
-uvicorn[standard]>=0.24.0
+uvicorn[standard]==0.42.0
-httpx>=0.25.0
+httpx==0.28.1
-qdrant-client>=1.6.0
+qdrant-client==1.17.1
-ollama>=0.1.0
+ollama==0.6.1
-toml>=0.10.2
+tiktoken==0.12.0
-tiktoken>=0.5.0
+apscheduler==3.11.2
-apscheduler>=3.10.0
+portalocker==3.2.0
-pytest>=7.0.0
+pytest==9.0.2
-pytest-asyncio>=0.21.0
+pytest-asyncio==1.3.0
-pytest-cov>=4.0.0
+pytest-cov==7.1.0

tests/conftest.py

@@ -0,0 +1,62 @@
+"""Shared test fixtures using production-realistic data."""
+import pytest
+from unittest.mock import AsyncMock, MagicMock, patch
+from app.config import Config
+
+
+@pytest.fixture
+def production_config():
+    """Config matching production deployment on deb8."""
+    config = MagicMock(spec=Config)
+    config.ollama_host = "http://10.0.0.10:11434"
+    config.qdrant_host = "http://10.0.0.22:6333"
+    config.qdrant_collection = "memories"
+    config.embedding_model = "snowflake-arctic-embed2"
+    config.semantic_token_budget = 25000
+    config.context_token_budget = 22000
+    config.semantic_search_turns = 2
+    config.semantic_score_threshold = 0.6
+    config.run_time = "02:00"
+    config.curator_model = "gpt-oss:120b"
+    config.debug = False
+    config.vector_size = 1024
+    config.cloud = MagicMock()
+    config.cloud.enabled = False
+    config.cloud.models = {}
+    config.cloud.get_cloud_model.return_value = None
+    return config
+
+
+@pytest.fixture
+def sample_qdrant_raw_payload():
+    """Sample raw payload from production Qdrant."""
+    return {
+        "type": "raw",
+        "text": "User: only change settings, not models\nAssistant: Changed semantic_token_budget from 25000 to 30000\nTimestamp: 2026-03-27T12:50:37.451593Z",
+        "timestamp": "2026-03-27T12:50:37.451593Z",
+        "role": "qa",
+        "content": "User: only change settings, not models\nAssistant: Changed semantic_token_budget from 25000 to 30000\nTimestamp: 2026-03-27T12:50:37.451593Z"
+    }
+
+
+@pytest.fixture
+def sample_ollama_models():
+    """Model list from production Ollama."""
+    return {
+        "models": [
+            {
+                "name": "snowflake-arctic-embed2:latest",
+                "model": "snowflake-arctic-embed2:latest",
+                "modified_at": "2026-02-16T16:43:44Z",
+                "size": 1160296718,
+                "details": {"family": "bert", "parameter_size": "566.70M", "quantization_level": "F16"}
+            },
+            {
+                "name": "gpt-oss:120b",
+                "model": "gpt-oss:120b",
+                "modified_at": "2026-03-11T12:45:48Z",
+                "size": 65369818941,
+                "details": {"family": "gptoss", "parameter_size": "116.8B", "quantization_level": "MXFP4"}
+            }
+        ]
+    }

tests/test_curator.py

@@ -2,9 +2,9 @@
 import pytest
 import json
 import os
-from datetime import datetime, timedelta
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
-from unittest.mock import MagicMock, patch
+from unittest.mock import MagicMock, AsyncMock, patch
 
 
 def make_curator():
@@ -77,14 +77,14 @@ class TestIsRecent:
     def test_memory_within_window(self):
         """Memory timestamped 1 hour ago is recent (within 24h)."""
         curator, _ = make_curator()
-        ts = (datetime.utcnow() - timedelta(hours=1)).isoformat() + "Z"
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=1)).isoformat() + "Z"
         memory = {"timestamp": ts}
         assert curator._is_recent(memory, hours=24) is True
 
     def test_memory_outside_window(self):
         """Memory timestamped 48 hours ago is not recent."""
         curator, _ = make_curator()
-        ts = (datetime.utcnow() - timedelta(hours=48)).isoformat() + "Z"
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=48)).isoformat() + "Z"
         memory = {"timestamp": ts}
         assert curator._is_recent(memory, hours=24) is False
 
@@ -109,7 +109,7 @@ class TestIsRecent:
     def test_boundary_edge_just_inside(self):
         """Memory at exactly hours-1 minutes ago should be recent."""
         curator, _ = make_curator()
-        ts = (datetime.utcnow() - timedelta(hours=23, minutes=59)).isoformat() + "Z"
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=23, minutes=59)).isoformat() + "Z"
         memory = {"timestamp": ts}
         assert curator._is_recent(memory, hours=24) is True
 
@@ -197,3 +197,294 @@ class TestAppendRuleToFile:
         target = prompts_dir / "newfile.md"
         assert target.exists()
         assert "New rule here." in target.read_text()
+
+
+class TestFormatExistingMemories:
+    """Tests for Curator._format_existing_memories."""
+
+    def test_empty_list_returns_no_memories_message(self):
+        """Empty list returns a 'no memories' message."""
+        curator, _ = make_curator()
+        result = curator._format_existing_memories([])
+        assert "No existing curated memories" in result
+
+    def test_single_memory_formatted(self):
+        """Single memory text is included in output."""
+        curator, _ = make_curator()
+        memories = [{"text": "User: hello\nAssistant: hi there"}]
+        result = curator._format_existing_memories(memories)
+        assert "hello" in result
+        assert "hi there" in result
+
+    def test_limits_to_last_20(self):
+        """Only last 20 memories are included."""
+        curator, _ = make_curator()
+        memories = [{"text": f"memory {i}"} for i in range(30)]
+        result = curator._format_existing_memories(memories)
+        # Should contain memory 10-29 (last 20), not memory 0-9
+        assert "memory 29" in result
+        assert "memory 10" in result
+
+
+class TestCallLlm:
+    """Tests for Curator._call_llm."""
+
+    @pytest.mark.asyncio
+    async def test_call_llm_returns_response(self):
+        """_call_llm returns the response text from Ollama."""
+        curator, _ = make_curator()
+
+        mock_resp = MagicMock()
+        mock_resp.json.return_value = {"response": "some LLM output"}
+
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+        mock_client.post = AsyncMock(return_value=mock_resp)
+
+        with patch("httpx.AsyncClient", return_value=mock_client):
+            result = await curator._call_llm("test prompt")
+
+        assert result == "some LLM output"
+        call_kwargs = mock_client.post.call_args
+        assert "test-model" in call_kwargs[1]["json"]["model"]
+
+    @pytest.mark.asyncio
+    async def test_call_llm_returns_empty_on_error(self):
+        """_call_llm returns empty string when Ollama errors."""
+        curator, _ = make_curator()
+
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+        mock_client.post = AsyncMock(side_effect=Exception("connection refused"))
+
+        with patch("httpx.AsyncClient", return_value=mock_client):
+            result = await curator._call_llm("test prompt")
+
+        assert result == ""
+
+
+class TestCuratorRun:
+    """Tests for Curator.run() method."""
+
+    @pytest.mark.asyncio
+    async def test_run_no_raw_memories_exits_early(self):
+        """run() exits early when no raw memories found."""
+        curator, mock_qdrant = make_curator()
+
+        # Mock scroll to return no points
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([], None))
+        mock_qdrant.collection = "memories"
+
+        await curator.run()
+        # Should not call LLM since there are no raw memories
+        # If it got here without error, that's success
+
+    @pytest.mark.asyncio
+    async def test_run_processes_raw_memories(self):
+        """run() processes raw memories and stores curated results."""
+        curator, mock_qdrant = make_curator()
+
+        # Create mock points
+        mock_point = MagicMock()
+        mock_point.id = "point-1"
+        mock_point.payload = {
+            "type": "raw",
+            "text": "User: hello\nAssistant: hi",
+            "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+        }
+
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([mock_point], None))
+        mock_qdrant.collection = "memories"
+        mock_qdrant.store_turn = AsyncMock(return_value="new-id")
+        mock_qdrant.delete_points = AsyncMock()
+
+        llm_response = json.dumps({
+            "new_curated_turns": [{"content": "User: hello\nAssistant: hi"}],
+            "permanent_rules": [],
+            "deletions": [],
+            "summary": "Curated one turn"
+        })
+
+        with patch.object(curator, "_call_llm", AsyncMock(return_value=llm_response)):
+            await curator.run()
+
+        mock_qdrant.store_turn.assert_called_once()
+        mock_qdrant.delete_points.assert_called()
+
+    @pytest.mark.asyncio
+    async def test_run_monthly_mode_on_day_01(self):
+        """run() uses monthly mode on day 01, processing all raw memories."""
+        curator, mock_qdrant = make_curator()
+
+        # Create a mock point with an old timestamp (outside 24h window)
+        old_ts = (datetime.now(timezone.utc) - timedelta(hours=72)).isoformat().replace("+00:00", "Z")
+        mock_point = MagicMock()
+        mock_point.id = "old-point"
+        mock_point.payload = {
+            "type": "raw",
+            "text": "User: old question\nAssistant: old answer",
+            "timestamp": old_ts,
+        }
+
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([mock_point], None))
+        mock_qdrant.collection = "memories"
+        mock_qdrant.store_turn = AsyncMock(return_value="new-id")
+        mock_qdrant.delete_points = AsyncMock()
+
+        llm_response = json.dumps({
+            "new_curated_turns": [],
+            "permanent_rules": [],
+            "deletions": [],
+            "summary": "Nothing to curate"
+        })
+
+        # Mock day 01
+        mock_now = datetime(2026, 4, 1, 2, 0, 0, tzinfo=timezone.utc)
+        with patch.object(curator, "_call_llm", AsyncMock(return_value=llm_response)), \
+             patch("app.curator.datetime") as mock_dt:
+            mock_dt.now.return_value = mock_now
+            mock_dt.fromisoformat = datetime.fromisoformat
+            mock_dt.side_effect = lambda *args, **kw: datetime(*args, **kw)
+            await curator.run()
+
+        # In monthly mode, even old memories are processed, so LLM should be called
+        # and delete_points should be called for the raw memory
+        mock_qdrant.delete_points.assert_called()
+
+    @pytest.mark.asyncio
+    async def test_run_handles_permanent_rules(self):
+        """run() appends permanent rules to prompt files."""
+        curator, mock_qdrant = make_curator()
+
+        mock_point = MagicMock()
+        mock_point.id = "point-1"
+        mock_point.payload = {
+            "type": "raw",
+            "text": "User: remember this\nAssistant: ok",
+            "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+        }
+
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([mock_point], None))
+        mock_qdrant.collection = "memories"
+        mock_qdrant.store_turn = AsyncMock(return_value="new-id")
+        mock_qdrant.delete_points = AsyncMock()
+
+        llm_response = json.dumps({
+            "new_curated_turns": [],
+            "permanent_rules": [{"rule": "Always be concise.", "target_file": "systemprompt.md"}],
+            "deletions": [],
+            "summary": "Added a rule"
+        })
+
+        with patch.object(curator, "_call_llm", AsyncMock(return_value=llm_response)), \
+             patch.object(curator, "_append_rule_to_file", AsyncMock()) as mock_append:
+            await curator.run()
+
+        mock_append.assert_called_once_with("systemprompt.md", "Always be concise.")
+
+    @pytest.mark.asyncio
+    async def test_run_handles_deletions(self):
+        """run() deletes specified point IDs when they exist in the database."""
+        curator, mock_qdrant = make_curator()
+
+        mock_point = MagicMock()
+        mock_point.id = "point-1"
+        mock_point.payload = {
+            "type": "raw",
+            "text": "User: delete me\nAssistant: ok",
+            "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+        }
+
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([mock_point], None))
+        mock_qdrant.collection = "memories"
+        mock_qdrant.store_turn = AsyncMock(return_value="new-id")
+        mock_qdrant.delete_points = AsyncMock()
+
+        llm_response = json.dumps({
+            "new_curated_turns": [],
+            "permanent_rules": [],
+            "deletions": ["point-1"],
+            "summary": "Deleted one"
+        })
+
+        with patch.object(curator, "_call_llm", AsyncMock(return_value=llm_response)):
+            await curator.run()
+
+        # delete_points should be called at least twice: once for valid deletions, once for processed raw
+        assert mock_qdrant.delete_points.call_count >= 1
+
+    @pytest.mark.asyncio
+    async def test_run_handles_llm_parse_failure(self):
+        """run() handles LLM returning unparseable response gracefully."""
+        curator, mock_qdrant = make_curator()
+
+        mock_point = MagicMock()
+        mock_point.id = "point-1"
+        mock_point.payload = {
+            "type": "raw",
+            "text": "User: test\nAssistant: ok",
+            "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+        }
+
+        mock_qdrant.client = AsyncMock()
+        mock_qdrant.client.scroll = AsyncMock(return_value=([mock_point], None))
+        mock_qdrant.collection = "memories"
+
+        with patch.object(curator, "_call_llm", AsyncMock(return_value="not json at all!!!")):
+            # Should not raise - just return early
+            await curator.run()
+
+        # store_turn should NOT be called since parsing failed
+        mock_qdrant.store_turn = AsyncMock()
+        mock_qdrant.store_turn.assert_not_called()
+
+
+class TestLoadCuratorPrompt:
+    """Tests for load_curator_prompt function."""
+
+    def test_loads_from_prompts_dir(self, tmp_path):
+        """load_curator_prompt loads from PROMPTS_DIR."""
+        import app.curator as curator_module
+
+        prompts_dir = tmp_path / "prompts"
+        prompts_dir.mkdir()
+        (prompts_dir / "curator_prompt.md").write_text("Test curator prompt")
+
+        with patch.object(curator_module, "PROMPTS_DIR", prompts_dir):
+            from app.curator import load_curator_prompt
+            result = load_curator_prompt()
+
+        assert result == "Test curator prompt"
+
+    def test_falls_back_to_static_dir(self, tmp_path):
+        """load_curator_prompt falls back to STATIC_DIR."""
+        import app.curator as curator_module
+
+        prompts_dir = tmp_path / "prompts"  # does not exist
+        static_dir = tmp_path / "static"
+        static_dir.mkdir()
+        (static_dir / "curator_prompt.md").write_text("Static prompt")
+
+        with patch.object(curator_module, "PROMPTS_DIR", prompts_dir), \
+             patch.object(curator_module, "STATIC_DIR", static_dir):
+            from app.curator import load_curator_prompt
+            result = load_curator_prompt()
+
+        assert result == "Static prompt"
+
+    def test_raises_when_not_found(self, tmp_path):
+        """load_curator_prompt raises FileNotFoundError when file missing."""
+        import app.curator as curator_module
+
+        with patch.object(curator_module, "PROMPTS_DIR", tmp_path / "nope"), \
+             patch.object(curator_module, "STATIC_DIR", tmp_path / "also_nope"):
+            from app.curator import load_curator_prompt
+            with pytest.raises(FileNotFoundError):
+                load_curator_prompt()

tests/test_integration.py

@@ -349,3 +349,83 @@ class TestApiChatStreaming:
         # Response body should contain both chunks concatenated
         body_text = resp.text
         assert "Hello" in body_text or len(body_text) > 0
+
+
+# ---------------------------------------------------------------------------
+# Health check edge cases
+# ---------------------------------------------------------------------------
+
+class TestHealthCheckEdgeCases:
+    def test_health_ollama_timeout(self, app_with_mocks):
+        """GET / handles Ollama timeout gracefully."""
+        import httpx
+        from fastapi.testclient import TestClient
+
+        vera_app, _ = app_with_mocks
+
+        mock_client_instance = AsyncMock()
+        mock_client_instance.__aenter__ = AsyncMock(return_value=mock_client_instance)
+        mock_client_instance.__aexit__ = AsyncMock(return_value=False)
+        mock_client_instance.get = AsyncMock(side_effect=httpx.TimeoutException("timeout"))
+
+        with patch("httpx.AsyncClient", return_value=mock_client_instance):
+            with TestClient(vera_app, raise_server_exceptions=True) as client:
+                resp = client.get("/")
+
+        assert resp.status_code == 200
+        assert resp.json()["ollama"] == "unreachable"
+
+
+# ---------------------------------------------------------------------------
+# POST /curator/run
+# ---------------------------------------------------------------------------
+
+class TestTriggerCurator:
+    def test_trigger_curator_endpoint(self, app_with_mocks):
+        """POST /curator/run triggers curation and returns status."""
+        from fastapi.testclient import TestClient
+        import app.main as main_module
+
+        vera_app, _ = app_with_mocks
+
+        mock_curator = MagicMock()
+        mock_curator.run = AsyncMock()
+
+        with patch.object(main_module, "curator", mock_curator):
+            with TestClient(vera_app) as client:
+                resp = client.post("/curator/run")
+
+        assert resp.status_code == 200
+        assert resp.json()["status"] == "curation completed"
+        mock_curator.run.assert_called_once()
+
+
+# ---------------------------------------------------------------------------
+# Proxy catch-all
+# ---------------------------------------------------------------------------
+
+class TestProxyAll:
+    def test_non_chat_api_proxied(self, app_with_mocks):
+        """Non-chat API paths are proxied to Ollama."""
+        from fastapi.testclient import TestClient
+
+        vera_app, _ = app_with_mocks
+
+        async def fake_aiter_bytes():
+            yield b'{"status": "ok"}'
+
+        mock_resp = MagicMock()
+        mock_resp.status_code = 200
+        mock_resp.headers = {"content-type": "application/json"}
+        mock_resp.aiter_bytes = fake_aiter_bytes
+
+        mock_client_instance = AsyncMock()
+        mock_client_instance.__aenter__ = AsyncMock(return_value=mock_client_instance)
+        mock_client_instance.__aexit__ = AsyncMock(return_value=False)
+        mock_client_instance.request = AsyncMock(return_value=mock_resp)
+
+        with patch("httpx.AsyncClient", return_value=mock_client_instance):
+            with TestClient(vera_app) as client:
+                resp = client.get("/api/show")
+
+        assert resp.status_code == 200

tests/test_proxy_handler.py

@@ -219,3 +219,94 @@ class TestHandleChatNonStreaming:
         # The wrapper should be stripped
         assert "Memory context" not in stored_question
         assert "What is the answer?" in stored_question
+
+
+class TestDebugLog:
+    """Tests for debug_log function."""
+
+    def test_debug_log_writes_json_when_enabled(self, tmp_path):
+        """Debug log appends valid JSON line to file when debug=True."""
+        import json
+        from unittest.mock import patch, MagicMock
+
+        mock_config = MagicMock()
+        mock_config.debug = True
+
+        with patch("app.proxy_handler.config", mock_config), \
+             patch("app.proxy_handler.DEBUG_LOG_DIR", tmp_path):
+            from app.proxy_handler import debug_log
+            debug_log("test_cat", "test message", {"key": "value"})
+
+        log_files = list(tmp_path.glob("debug_*.log"))
+        assert len(log_files) == 1
+        content = log_files[0].read_text().strip()
+        entry = json.loads(content)
+        assert entry["category"] == "test_cat"
+        assert entry["message"] == "test message"
+        assert entry["data"]["key"] == "value"
+
+    def test_debug_log_skips_when_disabled(self, tmp_path):
+        """Debug log does nothing when debug=False."""
+        from unittest.mock import patch, MagicMock
+
+        mock_config = MagicMock()
+        mock_config.debug = False
+
+        with patch("app.proxy_handler.config", mock_config), \
+             patch("app.proxy_handler.DEBUG_LOG_DIR", tmp_path):
+            from app.proxy_handler import debug_log
+            debug_log("test_cat", "test message")
+
+        log_files = list(tmp_path.glob("debug_*.log"))
+        assert len(log_files) == 0
+
+    def test_debug_log_without_data(self, tmp_path):
+        """Debug log works without optional data parameter."""
+        import json
+        from unittest.mock import patch, MagicMock
+
+        mock_config = MagicMock()
+        mock_config.debug = True
+
+        with patch("app.proxy_handler.config", mock_config), \
+             patch("app.proxy_handler.DEBUG_LOG_DIR", tmp_path):
+            from app.proxy_handler import debug_log
+            debug_log("simple_cat", "no data here")
+
+        log_files = list(tmp_path.glob("debug_*.log"))
+        assert len(log_files) == 1
+        entry = json.loads(log_files[0].read_text().strip())
+        assert "data" not in entry
+        assert entry["category"] == "simple_cat"
+
+
+class TestForwardToOllama:
+    """Tests for forward_to_ollama function."""
+
+    @pytest.mark.asyncio
+    async def test_forwards_request_to_ollama(self):
+        """forward_to_ollama proxies request to Ollama host."""
+        from app.proxy_handler import forward_to_ollama
+        from unittest.mock import patch, AsyncMock, MagicMock
+
+        mock_request = AsyncMock()
+        mock_request.body = AsyncMock(return_value=b'{"model": "llama3"}')
+        mock_request.method = "POST"
+        mock_request.headers = {"content-type": "application/json", "content-length": "20"}
+
+        mock_resp = MagicMock()
+        mock_resp.status_code = 200
+
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+        mock_client.request = AsyncMock(return_value=mock_resp)
+
+        with patch("httpx.AsyncClient", return_value=mock_client):
+            result = await forward_to_ollama(mock_request, "/api/show")
+
+        assert result == mock_resp
+        mock_client.request.assert_called_once()
+        call_kwargs = mock_client.request.call_args
+        assert call_kwargs[1]["method"] == "POST"
+        assert "/api/show" in call_kwargs[1]["url"]

tests/test_qdrant_service.py

@@ -0,0 +1,256 @@
+"""Tests for QdrantService — all Qdrant and Ollama calls are mocked."""
+import pytest
+from unittest.mock import AsyncMock, MagicMock, patch
+
+
+def make_qdrant_service():
+    """Create a QdrantService with mocked AsyncQdrantClient."""
+    with patch("app.qdrant_service.AsyncQdrantClient") as MockClient:
+        mock_client = AsyncMock()
+        MockClient.return_value = mock_client
+
+        from app.qdrant_service import QdrantService
+        svc = QdrantService(
+            host="http://localhost:6333",
+            collection="test_memories",
+            embedding_model="snowflake-arctic-embed2",
+            vector_size=1024,
+            ollama_host="http://localhost:11434",
+        )
+
+    return svc, mock_client
+
+
+class TestEnsureCollection:
+    """Tests for _ensure_collection."""
+
+    @pytest.mark.asyncio
+    async def test_creates_collection_when_missing(self):
+        """Creates collection if it does not exist."""
+        svc, mock_client = make_qdrant_service()
+        mock_client.get_collection = AsyncMock(side_effect=Exception("not found"))
+        mock_client.create_collection = AsyncMock()
+
+        await svc._ensure_collection()
+
+        mock_client.create_collection.assert_called_once()
+        assert svc._collection_ensured is True
+
+    @pytest.mark.asyncio
+    async def test_skips_if_collection_exists(self):
+        """Does not create collection if it already exists."""
+        svc, mock_client = make_qdrant_service()
+        mock_client.get_collection = AsyncMock(return_value=MagicMock())
+        mock_client.create_collection = AsyncMock()
+
+        await svc._ensure_collection()
+
+        mock_client.create_collection.assert_not_called()
+        assert svc._collection_ensured is True
+
+    @pytest.mark.asyncio
+    async def test_skips_if_already_ensured(self):
+        """Skips entirely if _collection_ensured is True."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+        mock_client.get_collection = AsyncMock()
+
+        await svc._ensure_collection()
+
+        mock_client.get_collection.assert_not_called()
+
+
+class TestGetEmbedding:
+    """Tests for get_embedding."""
+
+    @pytest.mark.asyncio
+    async def test_returns_embedding_vector(self):
+        """Returns embedding from Ollama response."""
+        svc, _ = make_qdrant_service()
+        fake_embedding = [0.1] * 1024
+
+        mock_resp = MagicMock()
+        mock_resp.json.return_value = {"embedding": fake_embedding}
+
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+        mock_client.post = AsyncMock(return_value=mock_resp)
+
+        with patch("httpx.AsyncClient", return_value=mock_client):
+            result = await svc.get_embedding("test text")
+
+        assert result == fake_embedding
+        assert len(result) == 1024
+
+
+class TestStoreTurn:
+    """Tests for store_turn."""
+
+    @pytest.mark.asyncio
+    async def test_stores_raw_user_turn(self):
+        """Stores a user turn with proper payload."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+        mock_client.upsert = AsyncMock()
+
+        fake_embedding = [0.1] * 1024
+        with patch.object(svc, "get_embedding", AsyncMock(return_value=fake_embedding)):
+            point_id = await svc.store_turn(role="user", content="hello world")
+
+        assert isinstance(point_id, str)
+        mock_client.upsert.assert_called_once()
+        call_args = mock_client.upsert.call_args
+        point = call_args[1]["points"][0]
+        assert point.payload["type"] == "raw"
+        assert point.payload["role"] == "user"
+        assert "User: hello world" in point.payload["text"]
+
+    @pytest.mark.asyncio
+    async def test_stores_curated_turn(self):
+        """Stores a curated turn without role prefix in text."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+        mock_client.upsert = AsyncMock()
+
+        fake_embedding = [0.1] * 1024
+        with patch.object(svc, "get_embedding", AsyncMock(return_value=fake_embedding)):
+            point_id = await svc.store_turn(
+                role="curated",
+                content="User: q\nAssistant: a",
+                entry_type="curated"
+            )
+
+        call_args = mock_client.upsert.call_args
+        point = call_args[1]["points"][0]
+        assert point.payload["type"] == "curated"
+        # Curated text should be the content directly, not prefixed
+        assert point.payload["text"] == "User: q\nAssistant: a"
+
+    @pytest.mark.asyncio
+    async def test_stores_with_topic_and_metadata(self):
+        """Stores turn with optional topic and metadata."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+        mock_client.upsert = AsyncMock()
+
+        fake_embedding = [0.1] * 1024
+        with patch.object(svc, "get_embedding", AsyncMock(return_value=fake_embedding)):
+            await svc.store_turn(
+                role="assistant",
+                content="some response",
+                topic="python",
+                metadata={"source": "test"}
+            )
+
+        call_args = mock_client.upsert.call_args
+        point = call_args[1]["points"][0]
+        assert point.payload["topic"] == "python"
+        assert point.payload["source"] == "test"
+
+
+class TestStoreQaTurn:
+    """Tests for store_qa_turn."""
+
+    @pytest.mark.asyncio
+    async def test_stores_qa_turn(self):
+        """Stores a complete Q&A turn."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+        mock_client.upsert = AsyncMock()
+
+        fake_embedding = [0.1] * 1024
+        with patch.object(svc, "get_embedding", AsyncMock(return_value=fake_embedding)):
+            point_id = await svc.store_qa_turn("What is Python?", "A programming language.")
+
+        assert isinstance(point_id, str)
+        call_args = mock_client.upsert.call_args
+        point = call_args[1]["points"][0]
+        assert point.payload["type"] == "raw"
+        assert point.payload["role"] == "qa"
+        assert "What is Python?" in point.payload["text"]
+        assert "A programming language." in point.payload["text"]
+
+
+class TestSemanticSearch:
+    """Tests for semantic_search."""
+
+    @pytest.mark.asyncio
+    async def test_returns_matching_results(self):
+        """Returns formatted search results."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+
+        mock_point = MagicMock()
+        mock_point.id = "result-1"
+        mock_point.score = 0.85
+        mock_point.payload = {"text": "User: hello\nAssistant: hi", "type": "curated"}
+
+        mock_query_result = MagicMock()
+        mock_query_result.points = [mock_point]
+        mock_client.query_points = AsyncMock(return_value=mock_query_result)
+
+        fake_embedding = [0.1] * 1024
+        with patch.object(svc, "get_embedding", AsyncMock(return_value=fake_embedding)):
+            results = await svc.semantic_search("hello", limit=10, score_threshold=0.6)
+
+        assert len(results) == 1
+        assert results[0]["id"] == "result-1"
+        assert results[0]["score"] == 0.85
+        assert results[0]["payload"]["type"] == "curated"
+
+
+class TestGetRecentTurns:
+    """Tests for get_recent_turns."""
+
+    @pytest.mark.asyncio
+    async def test_returns_sorted_recent_turns(self):
+        """Returns turns sorted by timestamp descending."""
+        svc, mock_client = make_qdrant_service()
+        svc._collection_ensured = True
+
+        mock_point1 = MagicMock()
+        mock_point1.id = "old"
+        mock_point1.payload = {"timestamp": "2026-01-01T00:00:00Z", "text": "old turn"}
+
+        mock_point2 = MagicMock()
+        mock_point2.id = "new"
+        mock_point2.payload = {"timestamp": "2026-03-01T00:00:00Z", "text": "new turn"}
+
+        # OrderBy returns server-sorted results (newest first)
+        mock_client.scroll = AsyncMock(return_value=([mock_point2, mock_point1], None))
+
+        results = await svc.get_recent_turns(limit=2)
+
+        assert len(results) == 2
+        # Newest first (server-sorted via OrderBy)
+        assert results[0]["id"] == "new"
+        assert results[1]["id"] == "old"
+
+
+class TestDeletePoints:
+    """Tests for delete_points."""
+
+    @pytest.mark.asyncio
+    async def test_deletes_by_ids(self):
+        """Deletes points by their IDs."""
+        svc, mock_client = make_qdrant_service()
+        mock_client.delete = AsyncMock()
+
+        await svc.delete_points(["id1", "id2"])
+
+        mock_client.delete.assert_called_once()
+
+
+class TestClose:
+    """Tests for close."""
+
+    @pytest.mark.asyncio
+    async def test_closes_client(self):
+        """Closes the async Qdrant client."""
+        svc, mock_client = make_qdrant_service()
+        mock_client.close = AsyncMock()
+
+        await svc.close()
+
+        mock_client.close.assert_called_once()

tests/test_utils.py

@@ -1,6 +1,7 @@
 """Tests for utility functions."""
 import pytest
-from app.utils import count_tokens, truncate_by_tokens, parse_curated_turn
+from unittest.mock import AsyncMock, MagicMock, patch
+from app.utils import count_tokens, truncate_by_tokens, parse_curated_turn, build_augmented_messages, count_messages_tokens
 
 
 class TestCountTokens:
@@ -85,25 +86,95 @@ Assistant: Yes, very popular."""
         assert "Line 3" in result[0]["content"]
 
 
+class TestCountMessagesTokens:
+    """Tests for count_messages_tokens function."""
+
+    def test_empty_list(self):
+        """Empty message list returns 0."""
+        assert count_messages_tokens([]) == 0
+
+    def test_single_message(self):
+        """Single message counts tokens of its content."""
+        msgs = [{"role": "user", "content": "Hello world"}]
+        result = count_messages_tokens(msgs)
+        assert result > 0
+
+    def test_multiple_messages(self):
+        """Multiple messages sum up their token counts."""
+        msgs = [
+            {"role": "user", "content": "Hello"},
+            {"role": "assistant", "content": "Hi there, how can I help you today?"},
+        ]
+        result = count_messages_tokens(msgs)
+        assert result > count_messages_tokens([msgs[0]])
+
+    def test_message_without_content(self):
+        """Message without content field contributes 0 tokens."""
+        msgs = [{"role": "system"}]
+        assert count_messages_tokens(msgs) == 0
+
+
+class TestLoadSystemPrompt:
+    """Tests for load_system_prompt function."""
+
+    def test_loads_from_prompts_dir(self, tmp_path):
+        """Loads systemprompt.md from PROMPTS_DIR."""
+        import app.utils as utils_module
+
+        prompts_dir = tmp_path / "prompts"
+        prompts_dir.mkdir()
+        (prompts_dir / "systemprompt.md").write_text("You are Vera.")
+
+        with patch.object(utils_module, "PROMPTS_DIR", prompts_dir):
+            result = utils_module.load_system_prompt()
+
+        assert result == "You are Vera."
+
+    def test_falls_back_to_static_dir(self, tmp_path):
+        """Falls back to STATIC_DIR when PROMPTS_DIR has no file."""
+        import app.utils as utils_module
+
+        prompts_dir = tmp_path / "no_prompts"  # does not exist
+        static_dir = tmp_path / "static"
+        static_dir.mkdir()
+        (static_dir / "systemprompt.md").write_text("Static Vera.")
+
+        with patch.object(utils_module, "PROMPTS_DIR", prompts_dir), \
+             patch.object(utils_module, "STATIC_DIR", static_dir):
+            result = utils_module.load_system_prompt()
+
+        assert result == "Static Vera."
+
+    def test_returns_empty_when_not_found(self, tmp_path):
+        """Returns empty string when systemprompt.md not found anywhere."""
+        import app.utils as utils_module
+
+        with patch.object(utils_module, "PROMPTS_DIR", tmp_path / "nope"), \
+             patch.object(utils_module, "STATIC_DIR", tmp_path / "also_nope"):
+            result = utils_module.load_system_prompt()
+
+        assert result == ""
+
+
 class TestFilterMemoriesByTime:
     """Tests for filter_memories_by_time function."""
 
     def test_includes_recent_memory(self):
         """Memory with timestamp in the last 24h should be included."""
-        from datetime import datetime, timedelta
+        from datetime import datetime, timedelta, timezone
         from app.utils import filter_memories_by_time
 
-        ts = (datetime.utcnow() - timedelta(hours=1)).isoformat()
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=1)).isoformat()
         memories = [{"timestamp": ts, "text": "recent"}]
         result = filter_memories_by_time(memories, hours=24)
         assert len(result) == 1
 
     def test_excludes_old_memory(self):
         """Memory older than cutoff should be excluded."""
-        from datetime import datetime, timedelta
+        from datetime import datetime, timedelta, timezone
         from app.utils import filter_memories_by_time
 
-        ts = (datetime.utcnow() - timedelta(hours=48)).isoformat()
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=48)).isoformat()
         memories = [{"timestamp": ts, "text": "old"}]
         result = filter_memories_by_time(memories, hours=24)
         assert len(result) == 0
@@ -124,6 +195,16 @@ class TestFilterMemoriesByTime:
         result = filter_memories_by_time(memories, hours=24)
         assert len(result) == 1
 
+    def test_z_suffix_old_timestamp_excluded(self):
+        """Regression: chained .replace() was not properly handling Z suffix on old timestamps."""
+        from datetime import datetime, timedelta, timezone
+        from app.utils import filter_memories_by_time
+
+        old_ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=48)).isoformat() + "Z"
+        memories = [{"timestamp": old_ts, "text": "old with Z"}]
+        result = filter_memories_by_time(memories, hours=24)
+        assert len(result) == 0, f"Old Z-suffixed timestamp should be excluded but wasn't: {old_ts}"
+
     def test_empty_list(self):
         """Empty input returns empty list."""
         from app.utils import filter_memories_by_time
@@ -132,10 +213,10 @@ class TestFilterMemoriesByTime:
 
     def test_z_suffix_timestamp(self):
         """ISO timestamp with Z suffix should be handled correctly."""
-        from datetime import datetime, timedelta
+        from datetime import datetime, timedelta, timezone
         from app.utils import filter_memories_by_time
 
-        ts = (datetime.utcnow() - timedelta(hours=1)).isoformat() + "Z"
+        ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=1)).isoformat() + "Z"
         memories = [{"timestamp": ts, "text": "recent with Z"}]
         result = filter_memories_by_time(memories, hours=24)
         assert len(result) == 1
@@ -190,37 +271,6 @@ class TestMergeMemories:
         assert len(result["ids"]) == 2
 
 
-class TestCalculateTokenBudget:
-    """Tests for calculate_token_budget function."""
-
-    def test_default_ratios_sum(self):
-        """Default ratios should sum to 1.0 (system+semantic+context)."""
-        from app.utils import calculate_token_budget
-
-        result = calculate_token_budget(1000)
-        assert result["system"] + result["semantic"] + result["context"] == 1000
-
-    def test_custom_ratios(self):
-        """Custom ratios should produce correct proportional budgets."""
-        from app.utils import calculate_token_budget
-
-        result = calculate_token_budget(
-            100, system_ratio=0.1, semantic_ratio=0.6, context_ratio=0.3
-        )
-        assert result["system"] == 10
-        assert result["semantic"] == 60
-        assert result["context"] == 30
-
-    def test_zero_budget(self):
-        """Zero total budget yields all zeros."""
-        from app.utils import calculate_token_budget
-
-        result = calculate_token_budget(0)
-        assert result["system"] == 0
-        assert result["semantic"] == 0
-        assert result["context"] == 0
-
-
 class TestBuildAugmentedMessages:
     """Tests for build_augmented_messages function (mocked I/O)."""
 
@@ -316,4 +366,72 @@ class TestBuildAugmentedMessages:
             )
 
         contents = [m["content"] for m in result]
-        assert any("Old question" in c or "Old answer" in c for c in contents)
+        assert any("Old question" in c or "Old answer" in c for c in contents)
+
+    @pytest.mark.asyncio
+    async def test_system_prompt_appends_to_caller_system(self):
+        """systemprompt.md content appends to caller's system message."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value="Vera memory context"), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Hello"}
+            ]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["role"] == "system"
+            assert system_msg["content"] == "You are a helpful assistant.\n\nVera memory context"
+
+    @pytest.mark.asyncio
+    async def test_empty_system_prompt_passthrough(self):
+        """When systemprompt.md is empty, only caller's system message passes through."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value=""), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Hello"}
+            ]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["content"] == "You are a helpful assistant."
+
+    @pytest.mark.asyncio
+    async def test_no_caller_system_with_vera_prompt(self):
+        """When caller sends no system message but systemprompt.md exists, use vera prompt."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value="Vera memory context"), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [{"role": "user", "content": "Hello"}]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["role"] == "system"
+            assert system_msg["content"] == "Vera memory context"
+
+    @pytest.mark.asyncio
+    async def test_no_system_anywhere(self):
+        """When neither caller nor systemprompt.md provides system content, no system message."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value=""), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [{"role": "user", "content": "Hello"}]
+            result = await build_augmented_messages(incoming)
+
+            # First message should be user, not system
+            assert result[0]["role"] == "user"