diff --git a/.env b/.env
new file mode 100644
index 0000000..124cdd9
--- /dev/null
+++ b/.env
@@ -0,0 +1 @@
+ENV_DEBUG = true
\ No newline at end of file
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..9490572
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,41 @@
+# Python
+__pycache__/
+*.py[cod]
+*.pyo
+*.pyd
+*.egg
+*.egg-info/
+dist/
+build/
+wheels/
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# Poetry
+poetry.lock
+
+# pytest
+.pytest_cache/
+htmlcov/
+.coverage
+coverage.xml
+
+# mypy / type checkers
+.mypy_cache/
+.ruff_cache/
+
+# IDE
+.vscode/
+.idea/
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Agents
+AGENTS.md
+CLAUDE.md
+DESIGN_DOCUMENT_MODULE.md
\ No newline at end of file
diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..cbb8879
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,23 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+## [Unreleased]
+
+### Added
+- Project specification in `project.md` — architecture, API design, cache backend, metadata schema, logging strategy, and TODO for future features (JOIN, SELECT * support)
+- `.gitignore` for Python/Poetry project
+- `pyproject.toml` dependencies: `sqlglot`, `sqlalchemy`, `loguru`, `python-dotenv`; dev dependencies: `pytest`, `ruff`, `mypy`
+- `src/sqlmem/` package structure with src layout
+- `src/sqlmem/exceptions.py` — `ReadOnlyError` (blocks INSERT/UPDATE/DELETE), `UnsupportedQueryError` (blocks JOIN and SELECT *)
+- `src/sqlmem/config.py` — loads `.env`, configures `loguru` with DEBUG/INFO level based on `SQLMEM_DEBUG`
+- `src/sqlmem/_meta.py` — package version constant
+- `src/sqlmem/parser.py` — SQL Parser using `sqlglot`; extracts table and columns from SELECT, raises on writes/JOIN/wildcard
+- `src/sqlmem/registry.py` — Column Registry; accumulates requested columns per table, detects missing columns requiring re-fetch
+- `src/sqlmem/cache.py` — Cache Manager; SQLite in-memory storage, load from `cache.db` on startup (with schema version check), hourly backup thread, `atexit`/SIGTERM flush, metadata tables (`_sqlmem_meta`, `_sqlmem_tables`, `_sqlmem_columns`)
+- `src/sqlmem/executor.py` — Query Executor; cache hit/miss logic, re-fetch on new columns with WARNING log
+- `src/sqlmem/engine.py` — `CachingEngine` wrapper; public API compatible with SQLAlchemy, `invalidate(table)` for manual cache clearing
+- `src/sqlmem/__init__.py` — public exports: `CachingEngine`, `ReadOnlyError`, `UnsupportedQueryError`
+- `tests/test_parser.py` — parser tests: SELECT parsing, ReadOnlyError, UnsupportedQueryError
+- `tests/test_cache.py` — cache tests: load, data correctness, metadata, disk backup/reload
+- `tests/test_registry.py` — registry tests: accumulation, needs_refetch, table isolation
diff --git a/project.md b/project.md
new file mode 100644
index 0000000..1120387
--- /dev/null
+++ b/project.md
@@ -0,0 +1,202 @@
+# SQLmem — Project Specification
+
+## Cíl
+
+Python modul fungující jako **transparentní cache vrstva mezi SQLAlchemy a databází**. Aplikace volá SQLAlchemy stejně jako dosud — SQLmem sedí mezi nimi, zachytává SELECT dotazy a vrací výsledky z in-memory SQLite cache. Zápisy (INSERT/UPDATE/DELETE) jsou přepouštěny přímo do databáze bez zásahu.
+
+---
+
+## Architektura
+
+```
+Aplikace (SQLAlchemy)
+        │
+        ▼
+   [ SQLmem Proxy ]
+   ┌───────────────────────────────┐
+   │  SQL Parser                   │  → rozezná SELECT vs. zápis
+   │  Cache Manager (SQLite RAM)   │  → drží data v paměti
+   │  Query Executor               │  → cache hit / miss logika
+   └───────────────────────────────┘
+        │
+        ▼
+   Databáze (přes původní SQLAlchemy engine)
+```
+
+---
+
+## API
+
+Modul implementuje **SQLAlchemy event hooks / custom engine wrapper** — aplikace nemusí měnit způsob volání ani formát vrácených dat.
+
+```python
+from sqlmem import CachingEngine
+from sqlalchemy import create_engine
+
+base_engine = create_engine("postgresql://...")
+engine = CachingEngine(base_engine)
+
+# Dál se pracuje stejně jako s běžným SQLAlchemy enginem:
+with engine.connect() as conn:
+    result = conn.execute(text("SELECT * FROM users WHERE status = 'active'"))
+```
+
+`CachingEngine` vrací objekty kompatibilní s `CursorResult` — `fetchall()`, `fetchone()`, `keys()` atd. fungují stejně.
+
+---
+
+## Cache backend
+
+- **SQLite in-memory** jako primární úložiště — veškeré dotazy běží v RAM.
+- **Persistence na disk** (`cache.db`) ve třech situacích:
+  - **Při startu**: pokud soubor existuje, načte se do paměti (`ATTACH` + kopie).
+  - **Periodicky každou hodinu**: snapshot in-memory SQLite se zapíše na disk (backup API).
+  - **Při vypnutí**: finální zápis na disk před ukončením (signal handler + context manager).
+- Celé tabulky se při cache miss načtou z databáze a drží v paměti.
+
+---
+
+## Komponenty
+
+### 1. SQL Parser
+- Detekuje typ dotazu (SELECT / zápis).
+- Extrahuje názvy tabulek z FROM a JOIN klauzulí.
+- Extrahuje seznam požadovaných sloupců.
+- Detekuje `SELECT *` (wildcard) a JOIN — vyhodí `UnsupportedQueryError`.
+- Rozhoduje, zda je dotaz obsloužitelný z cache.
+
+### 2. Column Registry
+
+Modul se **za běhu učí**, jaké sloupce z každé tabulky aplikace potřebuje — nevyžaduje žádnou předem danou konfiguraci.
+
+**Logika při každém příchozím dotazu:**
+
+1. Parser detekuje `SELECT *` nebo JOIN → vyhodí `UnsupportedQueryError` (není implementováno).
+2. Parser extrahuje `(tabulka, sloupce)` z dotazu.
+3. Registry provede **union** nově požadovaných sloupců s již známými.
+4. Cache Manager zkontroluje, zda cache pro danou tabulku obsahuje všechny potřebné sloupce:
+   - **Ano** → dotaz jde přímo do SQLite RAM (cache hit).
+   - **Ne** → re-fetch tabulky z DB s rozšířenou sadou sloupců → přepíše cache → dotaz do SQLite RAM.
+
+**Příklad akumulace sloupců:**
+
+```
+Dotaz 1: SELECT A, B FROM T3   → Registry: T3 = {A, B}   → fetch T3(A,B) z DB
+Dotaz 2: SELECT A, D FROM T3   → Registry: T3 = {A, B, D} → re-fetch T3(A,B,D) z DB
+Dotaz 3: SELECT B FROM T3      → cache hit, žádný DB dotaz
+Dotaz 4: SELECT * FROM T3      → UnsupportedQueryError (wildcard není podporován)
+Dotaz 5: SELECT A FROM T3 JOIN T4 ... → UnsupportedQueryError (JOIN není podporován)
+```
+
+**Metadata tabulka `_sqlmem_columns`** (uložena v SQLite):
+
+```sql
+CREATE TABLE _sqlmem_columns (
+    table_name  TEXT NOT NULL,
+    column_name TEXT NOT NULL,
+    PRIMARY KEY (table_name, column_name)
+);
+```
+
+- Při startu se načte z `cache.db` — Registry ví, co bylo kešováno v minulé session.
+- Při každém rozšíření sady sloupců se záznam aktualizuje.
+
+---
+
+### 3. Cache Manager
+- Drží in-memory SQLite instanci.
+- Sleduje, které tabulky jsou již načteny.
+- Podporuje TTL (Time-to-Live) pro automatické vypršení cache.
+- Umožňuje manuální invalidaci konkrétní tabulky.
+- **Persistence**:
+  - Při inicializaci načte `cache.db` ze disku (pokud existuje) do paměti.
+  - Spustí background vlákno, které každou hodinu provede `sqlite3` backup API (`conn.backup(disk_conn)`).
+  - Zaregistruje `atexit` handler a `SIGTERM` handler pro finální zápis při vypnutí.
+- **Metadata tabulky** (`_sqlmem_meta`) uložená přímo v SQLite cache:
+
+```sql
+CREATE TABLE _sqlmem_meta (
+    key   TEXT PRIMARY KEY,
+    value TEXT NOT NULL
+);
+-- Záznamy:
+-- app_version       verze sqlmem balíčku (např. "0.1.0")
+-- schema_version    verze schématu cache.db (integer, pro migrace)
+-- created_at        ISO timestamp prvního vytvoření cache.db
+
+CREATE TABLE _sqlmem_tables (
+    table_name      TEXT PRIMARY KEY,
+    last_refresh_at TEXT NOT NULL,  -- ISO 8601 UTC timestamp
+    row_count       INTEGER
+);
+```
+
+  - `last_refresh_at` se aktualizuje pokaždé, když se tabulka znovu načte z databáze.
+  - Při načtení `cache.db` ze disku se zkontroluje `schema_version` — pokud nesedí, cache se zahodí a načte znovu z DB.
+
+### 3. Query Executor
+- **Cache hit**: Spustí dotaz přímo v in-memory SQLite, vrátí výsledek.
+- **Cache miss**: Načte potřebné tabulky z databáze → uloží do cache → spustí dotaz.
+- **Zápisy (INSERT / UPDATE / DELETE)**: Vyhodí výjimku `ReadOnlyError` a dotaz zablokuje. Modul je striktně read-only.
+
+---
+
+## Rozsah MVP
+
+- [ ] `CachingEngine` wrapper kompatibilní se SQLAlchemy
+- [ ] Načtení `cache.db` při startu do in-memory SQLite
+- [ ] Periodický zápis na disk každou hodinu (background vlákno)
+- [ ] Zápis na disk při vypnutí (`atexit` + `SIGTERM`)
+- [ ] Parser pro detekci tabulek, sloupců a typu dotazu
+- [ ] Column Registry s akumulační logikou (union sloupců, wildcard detekce)
+- [ ] Re-fetch cache při rozšíření sady sloupců
+- [ ] Cache Manager s SQLite in-memory backendem
+- [ ] Cache hit/miss logika v Query Executoru
+- [ ] TTL podpora na úrovni tabulky
+- [ ] Manuální invalidace cache (`engine.cache.invalidate("tabulka")`)
+- [ ] Testy pokrývající cache hit, cache miss a blokování zápisů (`ReadOnlyError`)
+
+---
+
+## Co modul NEŘEŠÍ (mimo scope)
+
+- INSERT/UPDATE/DELETE — tyto operace jsou **zakázány** a vyhodí `ReadOnlyError`
+- Redis nebo jiný distribuovaný backend
+- Transakční konzistence
+
+---
+
+## Logování
+
+Řízeno přes `loguru`. Úroveň se nastavuje v `.env`:
+
+```env
+SQLMEM_DEBUG=true   # DEBUG level — podrobný výpis každého dotazu, cache operace, backup
+                    # false (výchozí) — INFO a výše
+```
+
+| Level | Kdy |
+|---|---|
+| `DEBUG` | Každý příchozí dotaz, extrahované tabulky/sloupce, cache hit/miss/re-fetch, backup start/konec |
+| `INFO` | Start/stop modulu, načtení cache.db, periodický backup, refresh tabulky |
+| `WARNING` | Re-fetch tabulky kvůli novým sloupcům, blížící se TTL expirace |
+| `ERROR` | `ReadOnlyError`, `UnsupportedQueryError`, selhání připojení k DB, selhání zápisu cache.db |
+
+---
+
+## TODO — budoucí funkce
+
+- **Podpora `SELECT *` (wildcard)**: Načte celou tabulku do cache, označí ji jako `full` — další dotazy na libovolný sloupec jsou vždy cache hit bez re-fetch.
+- **Podpora JOIN**: Parser extrahuje sloupce z každé joinované tabulky zvlášť, Column Registry je sleduje nezávisle. Cache Manager zajistí, že všechny potřebné tabulky jsou v paměti před spuštěním dotazu.
+
+---
+
+## Technologie
+
+| Vrstva | Knihovna |
+|---|---|
+| SQL parsing | `sqlglot` |
+| Cache úložiště | `sqlite3` (stdlib) |
+| Integrace | SQLAlchemy events / engine wrapper |
+| Logování | `loguru`, `python-dotenv` |
+| Testování | `pytest` |
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..9ce3ea4
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,27 @@
+[project]
+name = "sqlmem"
+version = "0.1.0"
+description = ""
+authors = [
+    {name = "jan.doubravsky@gmail.com"}
+]
+readme = "README.md"
+requires-python = ">=3.14,<3.15"
+dependencies = [
+    "sqlglot (>=30.8.0,<31.0.0)",
+    "sqlalchemy (>=2.0.50,<3.0.0)",
+    "loguru (>=0.7.3,<0.8.0)",
+    "python-dotenv (>=1.2.2,<2.0.0)"
+]
+
+
+[build-system]
+requires = ["poetry-core>=2.0.0,<3.0.0"]
+build-backend = "poetry.core.masonry.api"
+
+[dependency-groups]
+dev = [
+    "pytest (>=9.0.3,<10.0.0)",
+    "ruff (>=0.15.15,<0.16.0)",
+    "mypy (>=2.1.0,<3.0.0)"
+]
diff --git a/src/sqlmem/__init__.py b/src/sqlmem/__init__.py
new file mode 100644
index 0000000..3e8a707
--- /dev/null
+++ b/src/sqlmem/__init__.py
@@ -0,0 +1,4 @@
+from .engine import CachingEngine
+from .exceptions import ReadOnlyError, UnsupportedQueryError
+
+__all__ = ["CachingEngine", "ReadOnlyError", "UnsupportedQueryError"]
diff --git a/src/sqlmem/_meta.py b/src/sqlmem/_meta.py
new file mode 100644
index 0000000..3dc1f76
--- /dev/null
+++ b/src/sqlmem/_meta.py
@@ -0,0 +1 @@
+__version__ = "0.1.0"
diff --git a/src/sqlmem/cache.py b/src/sqlmem/cache.py
new file mode 100644
index 0000000..00ba0c3
--- /dev/null
+++ b/src/sqlmem/cache.py
@@ -0,0 +1,158 @@
+import atexit
+import signal
+import sqlite3
+import threading
+from datetime import datetime, timezone
+from pathlib import Path
+
+from loguru import logger
+
+import sqlmem._meta as _meta
+
+SCHEMA_VERSION = 1
+
+
+class CacheManager:
+    def __init__(self, db_path: Path, backup_interval: int) -> None:
+        self._db_path = db_path
+        self._backup_interval = backup_interval
+        self._mem_conn = sqlite3.connect(":memory:", check_same_thread=False)
+        self._lock = threading.Lock()
+        self._closed = False
+
+        self._ensure_meta_tables()
+        self._load_from_disk()
+        self._start_backup_thread()
+
+        atexit.register(self._backup_to_disk)
+        signal.signal(signal.SIGTERM, self._on_sigterm)
+
+    @property
+    def connection(self) -> sqlite3.Connection:
+        return self._mem_conn
+
+    def _ensure_meta_tables(self) -> None:
+        self._mem_conn.executescript("""
+            CREATE TABLE IF NOT EXISTS _sqlmem_meta (
+                key   TEXT PRIMARY KEY,
+                value TEXT NOT NULL
+            );
+            CREATE TABLE IF NOT EXISTS _sqlmem_tables (
+                table_name      TEXT PRIMARY KEY,
+                last_refresh_at TEXT NOT NULL,
+                row_count       INTEGER
+            );
+            CREATE TABLE IF NOT EXISTS _sqlmem_columns (
+                table_name  TEXT NOT NULL,
+                column_name TEXT NOT NULL,
+                PRIMARY KEY (table_name, column_name)
+            );
+        """)
+        self._mem_conn.execute(
+            "INSERT OR IGNORE INTO _sqlmem_meta (key, value) VALUES (?, ?)",
+            ("app_version", _meta.__version__),
+        )
+        self._mem_conn.execute(
+            "INSERT OR IGNORE INTO _sqlmem_meta (key, value) VALUES (?, ?)",
+            ("schema_version", str(SCHEMA_VERSION)),
+        )
+        self._mem_conn.execute(
+            "INSERT OR IGNORE INTO _sqlmem_meta (key, value) VALUES (?, ?)",
+            ("created_at", _now()),
+        )
+        self._mem_conn.commit()
+
+    def _load_from_disk(self) -> None:
+        if not self._db_path.exists():
+            logger.info(f"No cache file found at {self._db_path}, starting fresh.")
+            return
+
+        logger.info(f"Loading cache from {self._db_path}")
+        disk_conn = sqlite3.connect(self._db_path)
+        try:
+            schema_version = disk_conn.execute(
+                "SELECT value FROM _sqlmem_meta WHERE key = 'schema_version'"
+            ).fetchone()
+            if schema_version is None or int(schema_version[0]) != SCHEMA_VERSION:
+                logger.warning("Cache schema version mismatch — discarding cache file, starting fresh.")
+                disk_conn.close()
+                return
+
+            disk_conn.backup(self._mem_conn)
+            logger.info("Cache loaded from disk successfully.")
+        except Exception as e:
+            logger.error(f"Failed to load cache from disk: {e} — starting fresh.")
+        finally:
+            disk_conn.close()
+
+    def _backup_to_disk(self) -> None:
+        if self._closed:
+            return
+        logger.info(f"Backing up cache to {self._db_path}")
+        try:
+            with self._lock:
+                disk_conn = sqlite3.connect(self._db_path)
+                self._mem_conn.backup(disk_conn)
+                disk_conn.close()
+            logger.info("Cache backup complete.")
+        except Exception as e:
+            logger.error(f"Cache backup failed: {e}")
+
+    def _start_backup_thread(self) -> None:
+        def loop() -> None:
+            event = threading.Event()
+            while not event.wait(self._backup_interval):
+                self._backup_to_disk()
+
+        t = threading.Thread(target=loop, daemon=True, name="sqlmem-backup")
+        t.start()
+        logger.debug(f"Backup thread started (interval={self._backup_interval}s)")
+
+    def _on_sigterm(self, signum, frame) -> None:
+        logger.info("SIGTERM received — flushing cache to disk.")
+        self._backup_to_disk()
+
+    def mark_table_refreshed(self, table: str, row_count: int) -> None:
+        with self._lock:
+            self._mem_conn.execute(
+                """
+                INSERT INTO _sqlmem_tables (table_name, last_refresh_at, row_count)
+                VALUES (?, ?, ?)
+                ON CONFLICT(table_name) DO UPDATE SET
+                    last_refresh_at = excluded.last_refresh_at,
+                    row_count = excluded.row_count
+                """,
+                (table, _now(), row_count),
+            )
+            self._mem_conn.commit()
+
+    def is_table_cached(self, table: str) -> bool:
+        row = self._mem_conn.execute(
+            "SELECT 1 FROM _sqlmem_tables WHERE table_name = ?", (table,)
+        ).fetchone()
+        return row is not None
+
+    def load_table(self, table: str, columns: list[str], source_conn: sqlite3.Connection) -> None:
+        cols = ", ".join(columns)
+        logger.info(f"Fetching {table!r} columns [{cols}] from source DB")
+        rows = source_conn.execute(f"SELECT {cols} FROM {table}").fetchall()
+
+        with self._lock:
+            self._mem_conn.execute(f"DROP TABLE IF EXISTS {table}")
+            col_defs = ", ".join(f"{c} TEXT" for c in columns)
+            self._mem_conn.execute(f"CREATE TABLE {table} ({col_defs})")
+            placeholders = ", ".join("?" * len(columns))
+            self._mem_conn.executemany(f"INSERT INTO {table} VALUES ({placeholders})", rows)
+            self._mem_conn.commit()
+
+        self.mark_table_refreshed(table, len(rows))
+        logger.info(f"Table {table!r} cached ({len(rows)} rows, columns: {columns})")
+
+    def close(self) -> None:
+        self._backup_to_disk()
+        self._closed = True
+        self._mem_conn.close()
+
+
+def _now() -> str:
+    return datetime.now(timezone.utc).isoformat()
diff --git a/src/sqlmem/config.py b/src/sqlmem/config.py
new file mode 100644
index 0000000..f8b7771
--- /dev/null
+++ b/src/sqlmem/config.py
@@ -0,0 +1,19 @@
+import os
+from pathlib import Path
+
+from dotenv import load_dotenv
+from loguru import logger
+
+load_dotenv()
+
+DEBUG = os.getenv("SQLMEM_DEBUG", "false").lower() == "true"
+CACHE_DB_PATH = Path(os.getenv("SQLMEM_CACHE_DB", "cache.db"))
+BACKUP_INTERVAL_SECONDS = int(os.getenv("SQLMEM_BACKUP_INTERVAL", "3600"))
+
+logger.remove()
+logger.add(
+    sink=lambda msg: print(msg, end=""),
+    level="DEBUG" if DEBUG else "INFO",
+    format="<green>{time:YYYY-MM-DD HH:mm:ss}</green> | <level>{level: <8}</level> | <cyan>{name}</cyan>:<cyan>{line}</cyan> - <level>{message}</level>",
+    colorize=True,
+)
diff --git a/src/sqlmem/engine.py b/src/sqlmem/engine.py
new file mode 100644
index 0000000..0325756
--- /dev/null
+++ b/src/sqlmem/engine.py
@@ -0,0 +1,43 @@
+import sqlite3
+
+from loguru import logger
+from sqlalchemy.engine import Engine
+
+from .cache import CacheManager
+from .config import BACKUP_INTERVAL_SECONDS, CACHE_DB_PATH
+from .executor import QueryExecutor
+from .parser import parse
+from .registry import ColumnRegistry
+
+
+class CachingEngine:
+    """Transparent SQLAlchemy-compatible cache layer."""
+
+    def __init__(self, source_engine: Engine) -> None:
+        self._source_engine = source_engine
+        self._cache = CacheManager(CACHE_DB_PATH, BACKUP_INTERVAL_SECONDS)
+        self._registry = ColumnRegistry(self._cache.connection)
+        logger.info("CachingEngine initialized.")
+
+    def execute(self, sql: str) -> list[dict]:
+        parsed = parse(sql)
+        with self._source_engine.connect() as sa_conn:
+            raw_conn: sqlite3.Connection = sa_conn.connection.dbapi_connection
+            executor = QueryExecutor(self._cache, self._registry, raw_conn)
+            return executor.execute(parsed)
+
+    def invalidate(self, table: str) -> None:
+        logger.info(f"Manually invalidating cache for table {table!r}")
+        with self._cache._lock:
+            self._cache.connection.execute(f"DROP TABLE IF EXISTS {table}")
+            self._cache.connection.execute(
+                "DELETE FROM _sqlmem_tables WHERE table_name = ?", (table,)
+            )
+            self._cache.connection.execute(
+                "DELETE FROM _sqlmem_columns WHERE table_name = ?", (table,)
+            )
+            self._cache.connection.commit()
+
+    def close(self) -> None:
+        self._cache.close()
+        logger.info("CachingEngine closed.")
diff --git a/src/sqlmem/exceptions.py b/src/sqlmem/exceptions.py
new file mode 100644
index 0000000..8a8bba4
--- /dev/null
+++ b/src/sqlmem/exceptions.py
@@ -0,0 +1,6 @@
+class ReadOnlyError(Exception):
+    """Raised when a write operation (INSERT/UPDATE/DELETE) is attempted."""
+
+
+class UnsupportedQueryError(Exception):
+    """Raised when a query uses unsupported features (JOIN, SELECT *)."""
diff --git a/src/sqlmem/executor.py b/src/sqlmem/executor.py
new file mode 100644
index 0000000..23c7ea4
--- /dev/null
+++ b/src/sqlmem/executor.py
@@ -0,0 +1,42 @@
+import sqlite3
+
+from loguru import logger
+
+from .cache import CacheManager
+from .parser import ParsedQuery
+from .registry import ColumnRegistry
+
+
+class QueryExecutor:
+    def __init__(self, cache: CacheManager, registry: ColumnRegistry, source_conn: sqlite3.Connection) -> None:
+        self._cache = cache
+        self._registry = registry
+        self._source_conn = source_conn
+
+    def execute(self, parsed: ParsedQuery) -> list[dict]:
+        table = parsed.table
+        columns = parsed.columns
+
+        missing = self._registry.needs_refetch(table, columns)
+        table_cached = self._cache.is_table_cached(table)
+
+        if missing or not table_cached:
+            if table_cached and missing:
+                logger.warning(
+                    f"Re-fetching {table!r} — new columns requested: {missing}. "
+                    f"Expanding cache from {self._registry.get_columns(table)} + {missing}"
+                )
+            all_columns = list(self._registry.get_columns(table)) + missing
+            self._cache.load_table(table, all_columns, self._source_conn)
+            self._registry.update(table, all_columns)
+        else:
+            logger.debug(f"Cache hit: {table!r} columns={columns}")
+
+        return self._run_in_memory(parsed)
+
+    def _run_in_memory(self, parsed: ParsedQuery) -> list[dict]:
+        logger.debug(f"Executing in SQLite RAM: {parsed.original_sql!r}")
+        cursor = self._cache.connection.execute(parsed.original_sql)
+        col_names = [desc[0] for desc in cursor.description]
+        rows = cursor.fetchall()
+        return [dict(zip(col_names, row)) for row in rows]
diff --git a/src/sqlmem/parser.py b/src/sqlmem/parser.py
new file mode 100644
index 0000000..2316654
--- /dev/null
+++ b/src/sqlmem/parser.py
@@ -0,0 +1,71 @@
+from dataclasses import dataclass
+
+import sqlglot
+import sqlglot.expressions as exp
+from loguru import logger
+
+from .exceptions import ReadOnlyError, UnsupportedQueryError
+
+WRITE_TYPES = (exp.Insert, exp.Update, exp.Delete)
+
+
+@dataclass
+class ParsedQuery:
+    table: str
+    columns: list[str]
+    original_sql: str
+
+
+def parse(sql: str) -> ParsedQuery:
+    logger.debug(f"Parsing SQL: {sql!r}")
+
+    statement = sqlglot.parse_one(sql)
+
+    if isinstance(statement, WRITE_TYPES):
+        raise ReadOnlyError(
+            f"Write operations are not allowed. Attempted: {type(statement).__name__.upper()}"
+        )
+
+    if not isinstance(statement, exp.Select):
+        raise UnsupportedQueryError(f"Only SELECT statements are supported, got: {sql!r}")
+
+    _check_joins(statement)
+    _check_wildcard(statement)
+
+    table = _extract_table(statement)
+    columns = _extract_columns(statement)
+
+    logger.debug(f"Parsed → table={table!r}, columns={columns}")
+    return ParsedQuery(table=table, columns=columns, original_sql=sql)
+
+
+def _check_joins(statement: exp.Select) -> None:
+    if statement.find(exp.Join):
+        raise UnsupportedQueryError("JOIN is not supported yet. Use simple single-table SELECT.")
+
+
+def _check_wildcard(statement: exp.Select) -> None:
+    for col in statement.find_all(exp.Column):
+        if isinstance(col.this, exp.Star):
+            raise UnsupportedQueryError("SELECT * is not supported yet. Specify columns explicitly.")
+    if statement.find(exp.Star):
+        raise UnsupportedQueryError("SELECT * is not supported yet. Specify columns explicitly.")
+
+
+def _extract_table(statement: exp.Select) -> str:
+    from_clause = statement.find(exp.From)
+    if not from_clause:
+        raise UnsupportedQueryError("SELECT without FROM is not supported.")
+    table = from_clause.find(exp.Table)
+    if not table:
+        raise UnsupportedQueryError("Could not extract table name from query.")
+    return table.name
+
+
+def _extract_columns(statement: exp.Select) -> list[str]:
+    columns = []
+    for col in statement.find_all(exp.Column):
+        columns.append(col.name)
+    if not columns:
+        raise UnsupportedQueryError("Could not extract column names from query.")
+    return columns
diff --git a/src/sqlmem/registry.py b/src/sqlmem/registry.py
new file mode 100644
index 0000000..ac89032
--- /dev/null
+++ b/src/sqlmem/registry.py
@@ -0,0 +1,48 @@
+import sqlite3
+from threading import Lock
+
+from loguru import logger
+
+
+class ColumnRegistry:
+    """Tracks which columns per table have been requested and are held in cache."""
+
+    def __init__(self, mem_conn: sqlite3.Connection) -> None:
+        self._conn = mem_conn
+        self._lock = Lock()
+        self._ensure_table()
+
+    def _ensure_table(self) -> None:
+        self._conn.execute("""
+            CREATE TABLE IF NOT EXISTS _sqlmem_columns (
+                table_name  TEXT NOT NULL,
+                column_name TEXT NOT NULL,
+                PRIMARY KEY (table_name, column_name)
+            )
+        """)
+        self._conn.commit()
+
+    def get_columns(self, table: str) -> set[str]:
+        rows = self._conn.execute(
+            "SELECT column_name FROM _sqlmem_columns WHERE table_name = ?", (table,)
+        ).fetchall()
+        return {row[0] for row in rows}
+
+    def needs_refetch(self, table: str, requested: list[str]) -> list[str]:
+        """Returns columns that are requested but not yet in registry (missing columns)."""
+        known = self.get_columns(table)
+        missing = [c for c in requested if c not in known]
+        return missing
+
+    def update(self, table: str, columns: list[str]) -> None:
+        with self._lock:
+            existing = self.get_columns(table)
+            new_columns = [c for c in columns if c not in existing]
+            if not new_columns:
+                return
+            self._conn.executemany(
+                "INSERT OR IGNORE INTO _sqlmem_columns (table_name, column_name) VALUES (?, ?)",
+                [(table, col) for col in new_columns],
+            )
+            self._conn.commit()
+            logger.info(f"Registry updated: {table!r} now tracks columns {self.get_columns(table)}")
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/test_cache.py b/tests/test_cache.py
new file mode 100644
index 0000000..dba7043
--- /dev/null
+++ b/tests/test_cache.py
@@ -0,0 +1,61 @@
+import sqlite3
+from pathlib import Path
+
+import pytest
+
+from sqlmem.cache import CacheManager
+
+
+@pytest.fixture
+def cache(tmp_path):
+    c = CacheManager(db_path=tmp_path / "test_cache.db", backup_interval=9999)
+    yield c
+    c.close()
+
+
+@pytest.fixture
+def source_conn():
+    conn = sqlite3.connect(":memory:")
+    conn.execute("CREATE TABLE users (name TEXT, email TEXT, status TEXT)")
+    conn.executemany(
+        "INSERT INTO users VALUES (?, ?, ?)",
+        [("alice", "alice@example.com", "active"), ("bob", "bob@example.com", "inactive")],
+    )
+    conn.commit()
+    yield conn
+    conn.close()
+
+
+def test_table_not_cached_initially(cache):
+    assert cache.is_table_cached("users") is False
+
+
+def test_load_table(cache, source_conn):
+    cache.load_table("users", ["name", "email"], source_conn)
+    assert cache.is_table_cached("users") is True
+
+
+def test_loaded_data_correct(cache, source_conn):
+    cache.load_table("users", ["name", "email"], source_conn)
+    rows = cache.connection.execute("SELECT name, email FROM users").fetchall()
+    assert len(rows) == 2
+    assert ("alice", "alice@example.com") in rows
+
+
+def test_mark_table_refreshed(cache, source_conn):
+    cache.load_table("users", ["name"], source_conn)
+    row = cache.connection.execute(
+        "SELECT row_count FROM _sqlmem_tables WHERE table_name = 'users'"
+    ).fetchone()
+    assert row[0] == 2
+
+
+def test_backup_and_reload(tmp_path, source_conn):
+    db_path = tmp_path / "cache.db"
+    c1 = CacheManager(db_path=db_path, backup_interval=9999)
+    c1.load_table("users", ["name"], source_conn)
+    c1.close()
+
+    c2 = CacheManager(db_path=db_path, backup_interval=9999)
+    assert c2.is_table_cached("users") is True
+    c2.close()
diff --git a/tests/test_parser.py b/tests/test_parser.py
new file mode 100644
index 0000000..3770fc0
--- /dev/null
+++ b/tests/test_parser.py
@@ -0,0 +1,43 @@
+import pytest
+
+from sqlmem.exceptions import ReadOnlyError, UnsupportedQueryError
+from sqlmem.parser import parse
+
+
+def test_simple_select():
+    result = parse("SELECT name, email FROM users WHERE status = 'active'")
+    assert result.table == "users"
+    # WHERE columns are also extracted — needed for in-memory SQLite filtering
+    assert {"name", "email"}.issubset(set(result.columns))
+    assert "status" in result.columns
+
+
+def test_multiple_columns():
+    result = parse("SELECT a, b, c FROM orders")
+    assert result.table == "orders"
+    assert set(result.columns) == {"a", "b", "c"}
+
+
+def test_insert_raises_readonly():
+    with pytest.raises(ReadOnlyError):
+        parse("INSERT INTO users (name) VALUES ('alice')")
+
+
+def test_update_raises_readonly():
+    with pytest.raises(ReadOnlyError):
+        parse("UPDATE users SET name = 'bob' WHERE id = 1")
+
+
+def test_delete_raises_readonly():
+    with pytest.raises(ReadOnlyError):
+        parse("DELETE FROM users WHERE id = 1")
+
+
+def test_wildcard_raises_unsupported():
+    with pytest.raises(UnsupportedQueryError):
+        parse("SELECT * FROM users")
+
+
+def test_join_raises_unsupported():
+    with pytest.raises(UnsupportedQueryError):
+        parse("SELECT a.name, b.title FROM users a JOIN orders b ON a.id = b.user_id")
diff --git a/tests/test_registry.py b/tests/test_registry.py
new file mode 100644
index 0000000..22a0a0a
--- /dev/null
+++ b/tests/test_registry.py
@@ -0,0 +1,45 @@
+import sqlite3
+
+import pytest
+
+from sqlmem.registry import ColumnRegistry
+
+
+@pytest.fixture
+def registry():
+    conn = sqlite3.connect(":memory:")
+    return ColumnRegistry(conn)
+
+
+def test_empty_registry(registry):
+    assert registry.get_columns("users") == set()
+
+
+def test_update_and_get(registry):
+    registry.update("users", ["name", "email"])
+    assert registry.get_columns("users") == {"name", "email"}
+
+
+def test_update_accumulates(registry):
+    registry.update("users", ["name", "email"])
+    registry.update("users", ["email", "status"])
+    assert registry.get_columns("users") == {"name", "email", "status"}
+
+
+def test_needs_refetch_missing(registry):
+    registry.update("users", ["name"])
+    missing = registry.needs_refetch("users", ["name", "email"])
+    assert missing == ["email"]
+
+
+def test_needs_refetch_none_missing(registry):
+    registry.update("users", ["name", "email"])
+    missing = registry.needs_refetch("users", ["name"])
+    assert missing == []
+
+
+def test_independent_tables(registry):
+    registry.update("users", ["name"])
+    registry.update("orders", ["total"])
+    assert registry.get_columns("users") == {"name"}
+    assert registry.get_columns("orders") == {"total"}