diff --git a/setup.py b/setup.py
index 1f0d1299..f97ccb4a 100644
--- a/setup.py
+++ b/setup.py
@@ -38,8 +38,7 @@ def main() -> None:
             'tzlocal',
             'more_itertools',
             'pytz',
-            'sqlalchemy', # DB api
-            'cachew>=0.8.0', # caching with type hints
+            'sqlalchemy>=2.0',  # DB api
 
             *DEPS_INDEXER,
             *DEPS_SERVER,
@@ -84,8 +83,6 @@ def main() -> None:
 ]
 
 DEPS_SOURCES = {
-    # TODO make cachew optional?
-    # althrough server uses it so not sure...
     ('optional', 'dependencies that bring some bells & whistles'): [
         'logzero', # pretty colored logging
         'python-magic', # better mimetype decetion
diff --git a/src/promnesia/__main__.py b/src/promnesia/__main__.py
index 6401cb76..1bad03b7 100644
--- a/src/promnesia/__main__.py
+++ b/src/promnesia/__main__.py
@@ -18,7 +18,7 @@
 from .misc import install_server
 from .common import Extractor, PathIsh, logger, get_tmpdir, DbVisit, Res
 from .common import Source, get_system_tz, user_config_file, default_config_path
-from .dump import visits_to_sqlite
+from .database.dump import visits_to_sqlite
 from .extract import extract_visits
 
 
diff --git a/src/promnesia/compare.py b/src/promnesia/compare.py
index 382b9ad6..ce5dd1c9 100755
--- a/src/promnesia/compare.py
+++ b/src/promnesia/compare.py
@@ -8,6 +8,7 @@
 
 
 from .common import DbVisit, Url, PathWithMtime # TODO ugh. figure out pythonpath
+from .database.load import row_to_db_visit
 
 # TODO include latest too?
 # from cconfig import ignore, filtered
@@ -139,10 +140,10 @@ def compare_files(*files: Path, log=True) -> Iterator[Tuple[str, DbVisit]]:
         this_dts = name[0: name.index('.')] # can't use stem due to multiple extensions..
 
         from promnesia.server import _get_stuff # TODO ugh
-        engine, binder, table = _get_stuff(PathWithMtime.make(f))
+        engine, table = _get_stuff(PathWithMtime.make(f))
 
         with engine.connect() as conn:
-            vis = [binder.from_row(row) for row in conn.execute(table.select())]  # type: ignore[var-annotated]
+            vis = [row_to_db_visit(row) for row in conn.execute(table.select())]  # type: ignore[var-annotated]
 
         if last is not None:
             between = f'{last_dts}:{this_dts}'
diff --git a/src/promnesia/config.py b/src/promnesia/config.py
index 8164cae5..9e4903e0 100644
--- a/src/promnesia/config.py
+++ b/src/promnesia/config.py
@@ -69,6 +69,8 @@ def sources(self) -> Iterable[Res[Source]]:
 
     @property
     def cache_dir(self) -> Optional[Path]:
+        # TODO we used to use this for cachew, but it's best to rely on HPI modules etc to cofigure this
+        # keeping just in case for now
         cd = self.CACHE_DIR
         cpath: Optional[Path]
         if cd is None:
diff --git a/src/promnesia/database/common.py b/src/promnesia/database/common.py
new file mode 100644
index 00000000..09e30ed2
--- /dev/null
+++ b/src/promnesia/database/common.py
@@ -0,0 +1,66 @@
+from datetime import datetime
+from typing import Sequence, Tuple
+
+from sqlalchemy import (
+    Column,
+    Integer,
+    Row,
+    String,
+)
+
+# TODO maybe later move DbVisit here completely?
+# kinda an issue that it's technically an "api" because hook in config can patch up DbVisit
+from ..common import DbVisit, Loc
+
+
+def get_columns() -> Sequence[Column]:
+    # fmt: off
+    res: Sequence[Column] = [
+        Column('norm_url'     , String()),
+        Column('orig_url'     , String()),
+        Column('dt'           , String()),
+        Column('locator_title', String()),
+        Column('locator_href' , String()),
+        Column('src'          , String()),
+        Column('context'      , String()),
+        Column('duration'     , Integer())
+    ]
+    # fmt: on
+    assert len(res) == len(DbVisit._fields) + 1  # +1 because Locator is 'flattened'
+    return res
+
+
+def db_visit_to_row(v: DbVisit) -> Tuple:
+    # ugh, very hacky...
+    # we want to make sure the resulting tuple only consists of simple types
+    # so we can use dbengine directly
+    dt_s = v.dt.isoformat()
+    row = (
+        v.norm_url,
+        v.orig_url,
+        dt_s,
+        v.locator.title,
+        v.locator.href,
+        v.src,
+        v.context,
+        v.duration,
+    )
+    return row
+
+
+def row_to_db_visit(row: Sequence) -> DbVisit:
+    (norm_url, orig_url, dt_s, locator_title, locator_href, src, context, duration) = row
+    dt_s = dt_s.split()[0]  # backwards compatibility: previously it could be a string separated with tz name
+    dt = datetime.fromisoformat(dt_s)
+    return DbVisit(
+        norm_url=norm_url,
+        orig_url=orig_url,
+        dt=dt,
+        locator=Loc(
+            title=locator_title,
+            href=locator_href,
+        ),
+        src=src,
+        context=context,
+        duration=duration,
+    )
diff --git a/src/promnesia/dump.py b/src/promnesia/database/dump.py
similarity index 85%
rename from src/promnesia/dump.py
rename to src/promnesia/database/dump.py
index 6eb688dc..3d6178f0 100644
--- a/src/promnesia/dump.py
+++ b/src/promnesia/database/dump.py
@@ -1,7 +1,7 @@
 from pathlib import Path
 import shutil
 import sqlite3
-from typing import Dict, Iterable, List, Optional, Set, Tuple
+from typing import Dict, Iterable, List, Optional, Set
 
 from more_itertools import chunked
 
@@ -17,9 +17,7 @@
 )
 from sqlalchemy.dialects import sqlite as dialect_sqlite
 
-from cachew import NTBinder  # TODO need to get rid of this
-
-from .common import (
+from ..common import (
     DbVisit,
     Loc,
     Res,
@@ -28,7 +26,8 @@
     get_tmpdir,
     now_tz,
 )
-from . import config
+from .common import get_columns, db_visit_to_row
+from .. import config
 
 
 # NOTE: I guess the main performance benefit from this is not creating too many tmp lists and avoiding overhead
@@ -54,10 +53,10 @@ def enable_wal(dbapi_con, con_record) -> None:
 
 # returns critical warnings
 def visits_to_sqlite(
-        vit: Iterable[Res[DbVisit]],
-        *,
-        overwrite_db: bool,
-        _db_path: Optional[Path] = None,  # only used in tests
+    vit: Iterable[Res[DbVisit]],
+    *,
+    overwrite_db: bool,
+    _db_path: Optional[Path] = None,  # only used in tests
 ) -> List[Exception]:
     if _db_path is None:
         db_path = config.get().db
@@ -89,28 +88,8 @@ def vit_ok() -> Iterable[DbVisit]:
             index_stats[ev.src] = index_stats.get(ev.src, 0) + 1
             yield ev
 
-    binder = NTBinder.make(DbVisit)
     meta = MetaData()
-    # TODO is it ok to reuse meta/table??
-    table = Table('visits', meta, *binder.columns)
-
-    def db_visit_to_row(v: DbVisit) -> Tuple:
-        # ugh, very hacky...
-        # we want to make sure the resulting tuple only consists of simple types
-        # so we can use dbengine directly
-        dt = v.dt
-        dt_s = None if dt is None else dt.isoformat()
-        row = (
-            v.norm_url,
-            v.orig_url,
-            dt_s,
-            v.locator.title,
-            v.locator.href,
-            v.src,
-            v.context,
-            v.duration,
-        )
-        return row
+    table = Table('visits', meta, *get_columns())
 
     def query_total_stats(conn) -> Stats:
         query = select(table.c.src, func.count(table.c.src)).select_from(table).group_by(table.c.src)
diff --git a/src/promnesia/read_db.py b/src/promnesia/database/load.py
similarity index 69%
rename from src/promnesia/read_db.py
rename to src/promnesia/database/load.py
index 5ec72abb..8819175e 100644
--- a/src/promnesia/read_db.py
+++ b/src/promnesia/database/load.py
@@ -1,32 +1,29 @@
 from pathlib import Path
 from typing import Tuple, List
 
-from cachew import NTBinder
 from sqlalchemy import (
     create_engine,
     exc,
+    Engine,
     MetaData,
     Index,
     Table,
 )
-from sqlalchemy.engine import Engine
 
-from .common import DbVisit
+from .common import DbVisit, get_columns, row_to_db_visit
 
 
-DbStuff = Tuple[Engine, NTBinder, Table]
+DbStuff = Tuple[Engine, Table]
 
 
 def get_db_stuff(db_path: Path) -> DbStuff:
     assert db_path.exists(), db_path
     # todo how to open read only?
     # actually not sure if we can since we are creating an index here
-    engine = create_engine(f'sqlite:///{db_path}') # , echo=True)
-
-    binder = NTBinder.make(DbVisit)
+    engine = create_engine(f'sqlite:///{db_path}')  # , echo=True)
 
     meta = MetaData()
-    table = Table('visits', meta, *binder.columns)
+    table = Table('visits', meta, *get_columns())
 
     idx = Index('index_norm_url', table.c.norm_url)
     try:
@@ -39,13 +36,13 @@ def get_db_stuff(db_path: Path) -> DbStuff:
             raise e
 
     # NOTE: apparently it's ok to open connection on every request? at least my comparisons didn't show anything
-    return engine, binder, table
+    return engine, table
 
 
 def get_all_db_visits(db_path: Path) -> List[DbVisit]:
     # NOTE: this is pretty inefficient if the DB is huge
     # mostly intended for tests
-    engine, binder, table = get_db_stuff(db_path)
+    engine, table = get_db_stuff(db_path)
     query = table.select()
     with engine.connect() as conn:
-        return [binder.from_row(row) for row in conn.execute(query)]
+        return [row_to_db_visit(row) for row in conn.execute(query)]
diff --git a/src/promnesia/server.py b/src/promnesia/server.py
index 17671305..707691e5 100644
--- a/src/promnesia/server.py
+++ b/src/promnesia/server.py
@@ -18,7 +18,7 @@
 
 import fastapi
 
-from sqlalchemy import MetaData, exists, literal, between, or_, and_, exc, select
+from sqlalchemy import literal, between, or_, and_, exc, select
 from sqlalchemy import Column, Table, func, types
 from sqlalchemy.sql.elements import ColumnElement
 from sqlalchemy.sql import text
@@ -26,6 +26,7 @@
 
 from .common import PathWithMtime, DbVisit, Url, setup_logger, default_output_dir, get_system_tz
 from .cannon import canonify
+from .database.load import DbStuff, get_db_stuff, row_to_db_visit
 
 
 Json = Dict[str, Any]
@@ -117,8 +118,6 @@ def get_db_path(check: bool=True) -> Path:
     return db
 
 
-from .read_db import DbStuff, get_db_stuff
-
 @lru_cache(1)
 # PathWithMtime aids lru_cache in reloading the sqlalchemy binder
 def _get_stuff(db_path: PathWithMtime) -> DbStuff:
@@ -134,7 +133,7 @@ def get_stuff(db_path: Optional[Path]=None) -> DbStuff: # TODO better name
 
 
 def db_stats(db_path: Path) -> Json:
-    engine, binder, table = get_stuff(db_path)
+    engine, table = get_stuff(db_path)
     query = select(func.count()).select_from(table)
     with engine.connect() as conn:
         total = list(conn.execute(query))[0][0]
@@ -165,7 +164,7 @@ def search_common(url: str, where: Where) -> VisitsResponse:
         url = original_url
     logger.info('normalised url: %s', url)
 
-    engine, binder, table = get_stuff()
+    engine, table = get_stuff()
 
     query = table.select().where(where(table=table, url=url))
     logger.debug('query: %s', query)
@@ -173,7 +172,7 @@ def search_common(url: str, where: Where) -> VisitsResponse:
     with engine.connect() as conn:
         try:
             # TODO make more defensive here
-            visits: List[DbVisit] = [binder.from_row(row) for row in conn.execute(query)]
+            visits: List[DbVisit] = [row_to_db_visit(row) for row in conn.execute(query)]
         except exc.OperationalError as e:
             if getattr(e, 'msg', None) == 'no such table: visits':
                 logger.warn('you may have to run indexer first!')
@@ -361,7 +360,7 @@ def visited(request: VisitedRequest) -> VisitedResponse:
     if len(snurls) == 0:
         return []
 
-    engine, binder, table = get_stuff()
+    engine, table = get_stuff()
 
     # sqlalchemy doesn't seem to support SELECT FROM (VALUES (...)) in its api
     # also doesn't support array binding...
@@ -389,7 +388,7 @@ def visited(request: VisitedRequest) -> VisitedResponse:
     # brings down large queries to 50ms...
     with engine.connect() as conn:
         res = list(conn.execute(query))
-        present: Dict[str, Any] = {row[0]: binder.from_row(row[1:]) for row in res}
+        present: Dict[str, Any] = {row[0]: row_to_db_visit(row[1:]) for row in res}
     results = []
     for nu in nurls:
         r = present.get(nu, None)
diff --git a/src/promnesia/sources/browser_legacy.py b/src/promnesia/sources/browser_legacy.py
index 1bfbd51e..1b343bba 100644
--- a/src/promnesia/sources/browser_legacy.py
+++ b/src/promnesia/sources/browser_legacy.py
@@ -9,8 +9,14 @@
 from ..common import PathIsh, Results, Visit, Loc, logger, Second, is_sqlite_db
 from .. import config
 
-# todo mcachew?
-from cachew import cachew
+try:
+    from cachew import cachew  # type: ignore[import-not-found]
+except ModuleNotFoundError as me:
+    if me.name != 'cachew':
+        raise me
+    # this module is legacy anyway, so just make it defensive
+    def cachew(*args, **kwargs):  # type: ignore[no-redef]
+        return lambda f: f
 
 
 def index(p: PathIsh) -> Results:
diff --git a/src/promnesia/sources/takeout_legacy.py b/src/promnesia/sources/takeout_legacy.py
index da1043cf..2bf7ec85 100644
--- a/src/promnesia/sources/takeout_legacy.py
+++ b/src/promnesia/sources/takeout_legacy.py
@@ -34,7 +34,15 @@ def index() -> Results:
 
 
 from more_itertools import unique_everseen
-from cachew import cachew
+
+try:
+    from cachew import cachew  # type: ignore[import-not-found]
+except ModuleNotFoundError as me:
+    if me.name != 'cachew':
+        raise me
+    # this module is legacy anyway, so just make it defensive
+    def cachew(*args, **kwargs):  # type: ignore[no-redef]
+        return lambda f: f
 
 
 # TODO use CPath? Could encapsulate a path within an archive *or* within a directory
diff --git a/src/promnesia/tests/test_db_dump.py b/src/promnesia/tests/test_db_dump.py
index ccba7bad..50c5db84 100644
--- a/src/promnesia/tests/test_db_dump.py
+++ b/src/promnesia/tests/test_db_dump.py
@@ -13,9 +13,10 @@
 import pytz
 
 
-from ..common import DbVisit, Loc, Res
-from ..dump import visits_to_sqlite
-from ..read_db import get_all_db_visits
+from ..common import Loc, Res
+from ..database.common import DbVisit
+from ..database.dump import visits_to_sqlite
+from ..database.load import get_all_db_visits
 from ..sqlite import sqlite_connection
 
 from .common import gc_control, running_on_ci
@@ -73,9 +74,6 @@ def test_one_visit(tmp_path: Path) -> None:
 
     assert sqlite_visit == {
         'context': None,
-        # NOTE: at the moment date is dumped like this because of cachew NTBinder
-        # however it's not really necessary for promnesia (and possibly results in a bit of performance hit)
-        # I think we could just convert to a format sqlite supports, just need to make it backwards compatible
         'dt': '2023-11-14T23:11:01+01:00',
         'duration': 123,
         'locator_href': 'https://whatever.com',
@@ -109,7 +107,8 @@ def test_read_db_visits(tmp_path: Path) -> None:
 );
 '''
         )
-        # this tz format might occur in databases that were created when promnesia was using cachew NTBinder
+        # this dt format (zone name after iso timestap) might occur in legacy databases
+        # (that were created when promnesia was using cachew NTBinder)
         conn.execute(
             '''
 INSERT INTO visits VALUES(
diff --git a/tests/integration_test.py b/tests/integration_test.py
index 8bc5c59c..a85f2543 100644
--- a/tests/integration_test.py
+++ b/tests/integration_test.py
@@ -10,7 +10,7 @@
 from common import under_ci, DATA, GIT_ROOT, promnesia_bin
 
 from promnesia.common import _is_windows, DbVisit
-from promnesia.read_db import get_all_db_visits
+from promnesia.database.load import get_all_db_visits
 
 
 def run_index(cfg: Path, *, update=False) -> None:
diff --git a/tests/server_test.py b/tests/server_test.py
index 5b175908..92f9c45c 100644
--- a/tests/server_test.py
+++ b/tests/server_test.py
@@ -64,6 +64,7 @@ def wserver(db: Optional[PathIsh]=None): # TODO err not sure what type should it
 @contextmanager
 def _test_helper(tmp_path):
     tdir = Path(tmp_path)
+    # TODO probably don't need this anymore?
     cache_dir = tdir / 'cache'
     cache_dir.mkdir()