diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 01b95b0..a5a6a88 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -25,7 +25,7 @@ repos:
       - id: debug-statements
       - id: double-quote-string-fixer
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.9.2
+    rev: v0.9.3
     hooks:
       - id: ruff
         args: [--fix, --exit-non-zero-on-fix]
@@ -53,7 +53,7 @@ repos:
             ^docs/source/conf.py$
           )
   - repo: https://github.com/codespell-project/codespell
-    rev: v2.3.0
+    rev: v2.4.0
     hooks:
       - id: codespell
         additional_dependencies: [".[toml]"]
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 70d9ea1..f04f23c 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -25,7 +25,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Removed
 
--
+- Remove third-party dependency `cachetools` by [@XuehaiPan](https://github.com/XuehaiPan) in [#147](https://github.com/XuehaiPan/nvitop/pull/147).
 
 ------
 
diff --git a/README.md b/README.md
index f8f0b25..1291ed0 100644
--- a/README.md
+++ b/README.md
@@ -115,7 +115,6 @@ An interactive NVIDIA-GPU process viewer and beyond, the one-stop solution for G
 - NVIDIA Management Library (NVML)
 - nvidia-ml-py
 - psutil
-- cachetools
 - termcolor
 - curses<sup>[*](#curses)</sup> (with `libncursesw`)
 
diff --git a/docs/source/api/caching.rst b/docs/source/api/caching.rst
new file mode 100644
index 0000000..b9f269b
--- /dev/null
+++ b/docs/source/api/caching.rst
@@ -0,0 +1,10 @@
+nvitop.caching module
+---------------------
+
+.. currentmodule:: nvitop
+
+.. autosummary::
+
+    ttl_cache
+
+.. autofunction:: nvitop.ttl_cache
diff --git a/docs/source/index.rst b/docs/source/index.rst
index a6e6085..12e9e64 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -179,6 +179,7 @@ Please refer to section `More than a Monitor <https://github.com/XuehaiPan/nvito
     api/libnvml
     api/libcuda
     api/libcudart
+    api/caching
     api/utils
     select
     callbacks
diff --git a/docs/source/spelling_wordlist.txt b/docs/source/spelling_wordlist.txt
index cb70c1e..c1726d1 100644
--- a/docs/source/spelling_wordlist.txt
+++ b/docs/source/spelling_wordlist.txt
@@ -155,3 +155,6 @@ api
 utils
 GpuStatsLogger
 hostname
+len
+maxsize
+reentrant
diff --git a/nvitop/__init__.py b/nvitop/__init__.py
index 646d6f8..dcddd4d 100644
--- a/nvitop/__init__.py
+++ b/nvitop/__init__.py
@@ -20,7 +20,17 @@ import sys
 
 from nvitop import api
 from nvitop.api import *  # noqa: F403
-from nvitop.api import collector, device, host, libcuda, libcudart, libnvml, process, utils
+from nvitop.api import (
+    caching,
+    collector,
+    device,
+    host,
+    libcuda,
+    libcudart,
+    libnvml,
+    process,
+    utils,
+)
 from nvitop.select import select_devices
 from nvitop.version import __version__
 
@@ -28,7 +38,7 @@ from nvitop.version import __version__
 __all__ = [*api.__all__, 'select_devices']
 
 # Add submodules to the top-level namespace
-for submodule in (collector, device, host, libcuda, libcudart, libnvml, process, utils):
+for submodule in (caching, collector, device, host, libcuda, libcudart, libnvml, process, utils):
     sys.modules[f'{__name__}.{submodule.__name__.rpartition(".")[-1]}'] = submodule
 
 # Remove the nvitop.select module from sys.modules
diff --git a/nvitop/api/__init__.py b/nvitop/api/__init__.py
index e04e899..ad1d3c4 100644
--- a/nvitop/api/__init__.py
+++ b/nvitop/api/__init__.py
@@ -16,7 +16,18 @@
 # ==============================================================================
 """The core APIs of nvitop."""
 
-from nvitop.api import collector, device, host, libcuda, libcudart, libnvml, process, utils
+from nvitop.api import (
+    caching,
+    collector,
+    device,
+    host,
+    libcuda,
+    libcudart,
+    libnvml,
+    process,
+    utils,
+)
+from nvitop.api.caching import ttl_cache
 from nvitop.api.collector import ResourceMetricCollector, collect_in_background, take_snapshots
 from nvitop.api.device import (
     CudaDevice,
@@ -76,6 +87,8 @@ __all__ = [
     'take_snapshots',
     'collect_in_background',
     'ResourceMetricCollector',
+    # nvitop.api.caching
+    'ttl_cache',
     # nvitop.api.utils
     'NA',
     'NaType',
diff --git a/nvitop/api/caching.py b/nvitop/api/caching.py
new file mode 100644
index 0000000..6f62b9e
--- /dev/null
+++ b/nvitop/api/caching.py
@@ -0,0 +1,279 @@
+# This file is part of nvitop, the interactive NVIDIA-GPU process viewer.
+#
+# Copyright 2021-2025 Xuehai Pan. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Caching utilities."""
+
+from __future__ import annotations
+
+import builtins
+import functools
+import time
+from threading import RLock
+from typing import TYPE_CHECKING, Any, NamedTuple, overload
+
+
+if TYPE_CHECKING:
+    from collections.abc import Callable, Hashable, Sized
+    from collections.abc import Set as AbstractSet
+    from typing import TypeVar
+    from typing_extensions import ParamSpec, Self
+
+    _P = ParamSpec('_P')
+    _T = TypeVar('_T')
+
+
+__all__ = ['ttl_cache']
+
+
+class _CacheInfo(NamedTuple):
+    """A named tuple representing the cache statistics."""
+
+    hits: int
+    misses: int
+    maxsize: int
+    currsize: int
+
+
+try:
+    from functools import _make_key
+except ImportError:
+
+    class _HashedSeq(list):
+        """This class guarantees that hash() will be called no more than once per element."""
+
+        __slots__ = ('__hashvalue',)
+
+        def __init__(
+            self,
+            seq: tuple[Any, ...],
+            hash: Callable[[Any], int] = builtins.hash,  # pylint: disable=redefined-builtin
+        ) -> None:
+            """Initialize the hashed sequence."""
+            self[:] = seq
+            self.__hashvalue = hash(seq)
+
+        def __hash__(self) -> int:  # type: ignore[override]
+            """Return the hash value of the hashed sequence."""
+            return self.__hashvalue
+
+    _KWD_MARK = object()
+
+    # pylint: disable-next=too-many-arguments
+    def _make_key(  # type: ignore[misc]
+        args: tuple[Hashable, ...],
+        kwds: dict[str, Hashable],
+        typed: bool,
+        *,
+        kwd_mark: tuple[object, ...] = (_KWD_MARK,),
+        fasttypes: AbstractSet[type] = frozenset({int, str}),
+        tuple: type[tuple] = builtins.tuple,  # pylint: disable=redefined-builtin
+        type: type[type] = builtins.type,  # pylint: disable=redefined-builtin
+        len: Callable[[Sized], int] = builtins.len,  # pylint: disable=redefined-builtin
+    ) -> Hashable:
+        """Make a cache key from optionally typed positional and keyword arguments."""
+        key = args
+        if kwds:
+            key += kwd_mark
+            for item in kwds.items():
+                key += item
+        if typed:
+            key += tuple(type(v) for v in args)
+            if kwds:
+                key += tuple(type(v) for v in kwds.values())
+        elif len(key) == 1 and type(key[0]) in fasttypes:
+            return key[0]
+        return _HashedSeq(key)
+
+
+class _TTLCacheLink:  # pylint: disable=too-few-public-methods
+    __slots__ = ('expires', 'key', 'next', 'prev', 'value')
+
+    # pylint: disable-next=too-many-arguments,too-many-positional-arguments
+    def __init__(
+        self,
+        prev: Self | None,
+        next: Self | None,  # pylint: disable=redefined-builtin
+        key: Hashable,
+        value: Any,
+        expires: float | None,
+    ) -> None:
+        self.prev: Self = prev  # type: ignore[assignment]
+        self.next: Self = next  # type: ignore[assignment]
+        self.key: Hashable = key
+        self.value: Any = value
+        self.expires: float = expires  # type: ignore[assignment]
+
+
+@overload
+def ttl_cache(
+    maxsize: int | None = 128,
+    ttl: float = 600.0,
+    timer: Callable[[], float] = time.monotonic,
+    typed: bool = False,
+) -> Callable[[Callable[_P, _T]], Callable[_P, _T]]: ...
+
+
+@overload
+def ttl_cache(
+    maxsize: Callable[_P, _T],
+    ttl: float = 600.0,
+    timer: Callable[[], float] = time.monotonic,
+    typed: bool = False,
+) -> Callable[_P, _T]: ...
+
+
+# pylint: disable-next=too-many-statements
+def ttl_cache(
+    maxsize: int | Callable[_P, _T] | None = 128,
+    ttl: float = 600.0,
+    timer: Callable[[], float] = time.monotonic,
+    typed: bool = False,
+) -> Callable[[Callable[_P, _T]], Callable[_P, _T]] | Callable[_P, _T]:
+    """Time aware cache decorator."""
+    if isinstance(maxsize, int):
+        # Negative maxsize is treated as 0
+        maxsize = max(0, maxsize)
+    elif callable(maxsize) and isinstance(typed, bool):
+        # The user_function was passed in directly via the maxsize argument
+        func, maxsize = maxsize, 128
+        return ttl_cache(maxsize, ttl=ttl, timer=timer, typed=typed)(func)
+    elif maxsize is not None:
+        raise TypeError('Expected first argument to be an integer, a callable, or None')
+
+    if ttl < 0.0:
+        raise ValueError('TTL must be a non-negative number')
+    if not callable(timer):
+        raise TypeError('Timer must be a callable')
+
+    if maxsize == 0 or maxsize is None:
+        return functools.lru_cache(maxsize=maxsize, typed=typed)  # type: ignore[return-value]
+
+    # pylint: disable-next=too-many-statements,too-many-locals
+    def wrapper(func: Callable[_P, _T]) -> Callable[_P, _T]:
+        cache: dict[Any, _TTLCacheLink] = {}
+        cache_get = cache.get  # bound method to lookup a key or return None
+        cache_len = cache.__len__  # get cache size without calling len()
+        lock = RLock()  # because linked-list updates aren't thread-safe
+        root = _TTLCacheLink(*((None,) * 5))  # root of the circular doubly linked list
+        root.prev = root.next = root  # initialize by pointing to self
+        hits = misses = 0
+        full = False
+
+        def unlink(link: _TTLCacheLink) -> _TTLCacheLink:
+            with lock:
+                link_prev, link_next = link.prev, link.next
+                link_next.prev, link_prev.next = link_prev, link_next
+            return link_next
+
+        def append(link: _TTLCacheLink) -> _TTLCacheLink:
+            with lock:
+                last = root.prev
+                last.next = root.prev = link
+                link.prev, link.next = last, root
+            return link
+
+        def move_to_end(link: _TTLCacheLink) -> _TTLCacheLink:
+            with lock:
+                unlink(link)
+                append(link)
+            return link
+
+        def expire() -> None:
+            nonlocal full
+
+            with lock:
+                now = timer()
+                front = root.next
+                while front is not root and front.expires < now:
+                    del cache[front.key]
+                    front = unlink(front)
+                full = cache_len() >= maxsize
+
+        @functools.wraps(func)
+        def wrapped(*args: _P.args, **kwargs: _P.kwargs) -> _T:
+            # Size limited time aware caching
+            nonlocal root, hits, misses, full
+
+            key = _make_key(args, kwargs, typed)
+            with lock:
+                link = cache_get(key)
+                if link is not None:
+                    if timer() < link.expires:
+                        hits += 1
+                        return link.value
+                    expire()
+
+            misses += 1
+            result = func(*args, **kwargs)
+            expires = timer() + ttl
+            with lock:
+                if key in cache:
+                    # Getting here means that this same key was added to the cache while the lock
+                    # was released or the key was expired. Move the link to the front of the
+                    # circular queue.
+                    link = move_to_end(cache[key])
+                    # We need only update the expiration time.
+                    link.value = result
+                    link.expires = expires
+                else:
+                    if full:
+                        expire()
+                    if full:
+                        # Use the old root to store the new key and result.
+                        root.key = key
+                        root.value = result
+                        root.expires = expires
+                        # Empty the oldest link and make it the new root.
+                        # Keep a reference to the old key and old result to prevent their ref counts
+                        # from going to zero during the update. That will prevent potentially
+                        # arbitrary object clean-up code (i.e. __del__) from running while we're
+                        # still adjusting the links.
+                        front = root.next
+                        old_key = front.key
+                        front.key = front.value = front.expires = None  # type: ignore[assignment]
+                        # Now update the cache dictionary.
+                        del cache[old_key]
+                        # Save the potentially reentrant cache[key] assignment for last, after the
+                        # root and links have been put in a consistent state.
+                        cache[key], root = root, front
+                    else:
+                        # Put result in a new link at the front of the queue.
+                        cache[key] = append(_TTLCacheLink(None, None, key, result, expires))
+                    full = cache_len() >= maxsize
+            return result
+
+        def cache_info() -> _CacheInfo:
+            """Report cache statistics."""
+            with lock:
+                expire()
+                return _CacheInfo(hits, misses, maxsize, cache_len())
+
+        def cache_clear() -> None:
+            """Clear the cache and cache statistics."""
+            nonlocal hits, misses, full
+            with lock:
+                cache.clear()
+                root.prev = root.next = root
+                root.key = root.value = root.expires = None  # type: ignore[assignment]
+                hits = misses = 0
+                full = False
+
+        wrapped.cache_info = cache_info  # type: ignore[attr-defined]
+        wrapped.cache_clear = cache_clear  # type: ignore[attr-defined]
+        wrapped.cache_parameters = lambda: {'maxsize': maxsize, 'typed': typed}  # type: ignore[attr-defined]
+        return wrapped
+
+    return wrapper
diff --git a/nvitop/gui/library/__init__.py b/nvitop/gui/library/__init__.py
index 1c1c98e..9dba057 100644
--- a/nvitop/gui/library/__init__.py
+++ b/nvitop/gui/library/__init__.py
@@ -39,5 +39,6 @@ from nvitop.gui.library.utils import (
     cut_string,
     make_bar,
     set_color,
+    ttl_cache,
 )
 from nvitop.gui.library.widestring import WideString, wcslen
diff --git a/nvitop/gui/library/device.py b/nvitop/gui/library/device.py
index d29cfc8..a147b80 100644
--- a/nvitop/gui/library/device.py
+++ b/nvitop/gui/library/device.py
@@ -3,9 +3,7 @@
 
 # pylint: disable=missing-module-docstring,missing-class-docstring,missing-function-docstring
 
-from cachetools.func import ttl_cache
-
-from nvitop.api import NA, libnvml, utilization2string
+from nvitop.api import NA, libnvml, ttl_cache, utilization2string
 from nvitop.api import MigDevice as MigDeviceBase
 from nvitop.api import PhysicalDevice as DeviceBase
 from nvitop.gui.library.process import GpuProcess
diff --git a/nvitop/gui/library/process.py b/nvitop/gui/library/process.py
index 05d6a94..e12fbd0 100644
--- a/nvitop/gui/library/process.py
+++ b/nvitop/gui/library/process.py
@@ -3,7 +3,6 @@
 
 # pylint: disable=missing-module-docstring,missing-class-docstring,missing-function-docstring
 
-
 from nvitop.api import (
     NA,
     GiB,
diff --git a/nvitop/gui/library/utils.py b/nvitop/gui/library/utils.py
index 5ab64a7..d97ea48 100644
--- a/nvitop/gui/library/utils.py
+++ b/nvitop/gui/library/utils.py
@@ -7,10 +7,25 @@ import contextlib
 import math
 import os
 
-from nvitop.api import NA, colored, host, set_color  # noqa: F401 # pylint: disable=unused-import
+from nvitop.api import NA, colored, host, set_color, ttl_cache
 from nvitop.gui.library.widestring import WideString
 
 
+__all__ = [
+    'NA',
+    'USERNAME',
+    'HOSTNAME',
+    'SUPERUSER',
+    'USERCONTEXT',
+    'LARGE_INTEGER',
+    'ttl_cache',
+    'colored',
+    'set_color',
+    'cut_string',
+    'make_bar',
+]
+
+
 USERNAME = 'N/A'
 with contextlib.suppress(ImportError, OSError):
     USERNAME = host.getuser()
diff --git a/nvitop/gui/screens/main/device.py b/nvitop/gui/screens/main/device.py
index c47babb..fddeaff 100644
--- a/nvitop/gui/screens/main/device.py
+++ b/nvitop/gui/screens/main/device.py
@@ -6,9 +6,16 @@
 import threading
 import time
 
-from cachetools.func import ttl_cache
-
-from nvitop.gui.library import NA, Device, Displayable, colored, cut_string, host, make_bar
+from nvitop.gui.library import (
+    NA,
+    Device,
+    Displayable,
+    colored,
+    cut_string,
+    host,
+    make_bar,
+    ttl_cache,
+)
 from nvitop.version import __version__
 
 
diff --git a/nvitop/gui/screens/main/process.py b/nvitop/gui/screens/main/process.py
index d5e9c34..d91aee8 100644
--- a/nvitop/gui/screens/main/process.py
+++ b/nvitop/gui/screens/main/process.py
@@ -11,8 +11,6 @@ import time
 from operator import attrgetter, xor
 from typing import TYPE_CHECKING, Any, NamedTuple
 
-from cachetools.func import ttl_cache
-
 from nvitop.gui.library import (
     HOSTNAME,
     LARGE_INTEGER,
@@ -27,6 +25,7 @@ from nvitop.gui.library import (
     colored,
     cut_string,
     host,
+    ttl_cache,
     wcslen,
 )
 
diff --git a/nvitop/gui/screens/treeview.py b/nvitop/gui/screens/treeview.py
index e6a9f9b..cb7a5cd 100644
--- a/nvitop/gui/screens/treeview.py
+++ b/nvitop/gui/screens/treeview.py
@@ -9,8 +9,6 @@ from collections import deque
 from functools import partial
 from itertools import islice
 
-from cachetools.func import ttl_cache
-
 from nvitop.gui.library import (
     NA,
     SUPERUSER,
@@ -22,6 +20,7 @@ from nvitop.gui.library import (
     WideString,
     host,
     send_signal,
+    ttl_cache,
 )
 
 
diff --git a/pyproject.toml b/pyproject.toml
index 539651a..a684bf2 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -49,7 +49,6 @@ dependencies = [
     # Sync with nvitop/version.py and requirements.txt
     "nvidia-ml-py >= 11.450.51, < 12.561.0a0",
     "psutil >= 5.6.6",
-    "cachetools >= 1.0.1",
     "termcolor >= 1.0.0",
     "colorama >= 0.4.0; platform_system == 'Windows'",
     "windows-curses >= 2.2.0; platform_system == 'Windows'",
@@ -204,7 +203,6 @@ ignore = [
 
 [tool.ruff.lint.isort]
 known-first-party = ["nvitop", "nvitop_exporter"]
-known-local-folder = ["nvitop", "nvitop-exporter"]
 extra-standard-library = ["typing_extensions"]
 lines-after-imports = 2
 
diff --git a/requirements.txt b/requirements.txt
index b5bb43c..87929fa 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,7 +1,6 @@
 # Sync with pyproject.toml and nvitop/version.py
 nvidia-ml-py >= 11.450.51, < 12.561.0a0
 psutil >= 5.6.6
-cachetools >= 1.0.1
 termcolor >= 1.0.0
 colorama >= 0.4.0; platform_system == 'Windows'
 windows-curses >= 2.2.0; platform_system == 'Windows'