diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 381f2bdf9..baccccee5 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -20,11 +20,11 @@ repos:
       - id: black
         language_version: python3
   - repo: https://github.com/pre-commit/mirrors-mypy
-    rev: "v1.3.0"
+    rev: v1.8.0
     hooks:
       - id: mypy
         args: [--pretty, --ignore-missing-imports]
-        additional_dependencies: [types-requests]
+        additional_dependencies: [types-requests, types-setuptools]
   - repo: https://github.com/PyCQA/isort
     rev: "5.12.0"
     hooks:
diff --git a/dev-requirements.txt b/dev-requirements.txt
index c56d05280..11052e634 100644
--- a/dev-requirements.txt
+++ b/dev-requirements.txt
@@ -5,6 +5,7 @@ pyzotero
 python-dotenv
 pymupdf
 build
+types-setuptools
 types-requests
 langchain_openai
 langchain_community
diff --git a/paperqa/contrib/zotero.py b/paperqa/contrib/zotero.py
index 1d4330dc9..7640e499f 100644
--- a/paperqa/contrib/zotero.py
+++ b/paperqa/contrib/zotero.py
@@ -128,7 +128,7 @@ def get_pdf(self, item: dict) -> Union[Path, None]:
         if pdf_key is None:
             return None
 
-        pdf_path: Path = Path(self.storage / (pdf_key + ".pdf"))  # type: ignore
+        pdf_path: Path = Path(self.storage / (pdf_key + ".pdf"))  # type: ignore[operator]
 
         if not pdf_path.exists():
             pdf_path.parent.mkdir(parents=True, exist_ok=True)
diff --git a/paperqa/docs.py b/paperqa/docs.py
index acace0b04..a70f38f9a 100644
--- a/paperqa/docs.py
+++ b/paperqa/docs.py
@@ -422,7 +422,7 @@ async def aadd(
             # get first name and year from citation
             match = re.search(r"([A-Z][a-z]+)", citation)
             if match is not None:
-                author = match.group(1)  # type: ignore
+                author = match.group(1)
             else:
                 # panicking - no word??
                 raise ValueError(
@@ -433,7 +433,7 @@ async def aadd(
             year = ""
             match = re.search(r"(\d{4})", citation)
             if match is not None:
-                year = match.group(1)  # type: ignore
+                year = match.group(1)
             docname = f"{author}{year}"
         docname = self._get_unique_name(docname)
         doc = Doc(docname=docname, citation=citation, dockey=dockey)
@@ -619,7 +619,7 @@ async def aget_evidence(
     ) -> Answer:
         if len(self.docs) == 0 and self.docs_index is None:
             # do we have no docs?
-            return answer
+            return answer  # type: ignore[unreachable]
         self._build_texts_index(keys=answer.dockey_filter)
         _k = k
         if answer.dockey_filter is not None:
@@ -663,13 +663,13 @@ async def process(match):
                 score = 5
             else:
                 if self.prompts.json_summary:
-                    summary_chain = self.summary_llm_model.make_chain(
+                    summary_chain = self.summary_llm_model.make_chain(  # type: ignore[union-attr]
                         client=self._client,
                         prompt=self.prompts.summary_json,
                         system_prompt=self.prompts.summary_json_system,
                     )
                 else:
-                    summary_chain = self.summary_llm_model.make_chain(
+                    summary_chain = self.summary_llm_model.make_chain(  # type: ignore[union-attr]
                         client=self._client,
                         prompt=self.prompts.summary,
                         system_prompt=self.prompts.system,
diff --git a/paperqa/llms.py b/paperqa/llms.py
index d7c7fa182..1d70c4c5e 100644
--- a/paperqa/llms.py
+++ b/paperqa/llms.py
@@ -205,9 +205,9 @@ async def execute(
                         f for f in callbacks if not is_coroutine_callable(f)
                     ]
                     async_callbacks = [f for f in callbacks if is_coroutine_callable(f)]
-                    completion = self.achat_iter(client, messages)  # type: ignore
+                    completion = self.achat_iter(client, messages)
                     text_result = []
-                    async for chunk in completion:  # type: ignore
+                    async for chunk in completion:  # type: ignore[attr-defined]
                         if chunk:
                             if result.seconds_to_first_token == 0:
                                 result.seconds_to_first_token = (
@@ -250,12 +250,12 @@ async def execute(
                     ]
                     async_callbacks = [f for f in callbacks if is_coroutine_callable(f)]
 
-                    completion = self.acomplete_iter(  # type: ignore
+                    completion = self.acomplete_iter(
                         client,
                         formatted_prompt,
                     )
                     text_result = []
-                    async for chunk in completion:  # type: ignore
+                    async for chunk in completion:  # type: ignore[attr-defined]
                         if chunk:
                             if result.seconds_to_first_token == 0:
                                 result.seconds_to_first_token = (
@@ -289,7 +289,7 @@ def _check_client(self, client: Any) -> AsyncOpenAI:
             raise ValueError(
                 f"Your client is not a required AsyncOpenAI client. It is a {type(client)}"
             )
-        return cast(AsyncOpenAI, client)
+        return client
 
     @model_validator(mode="after")
     @classmethod
@@ -323,14 +323,14 @@ async def acomplete_iter(self, client: Any, prompt: str) -> Any:
     async def achat(self, client: Any, messages: list[dict[str, str]]) -> str:
         aclient = self._check_client(client)
         completion = await aclient.chat.completions.create(
-            messages=messages, **process_llm_config(self.config)  # type: ignore
+            messages=messages, **process_llm_config(self.config)
         )
         return completion.choices[0].message.content or ""
 
     async def achat_iter(self, client: Any, messages: list[dict[str, str]]) -> Any:
         aclient = self._check_client(client)
         completion = await aclient.chat.completions.create(
-            messages=messages, **process_llm_config(self.config), stream=True  # type: ignore
+            messages=messages, **process_llm_config(self.config), stream=True
         )
         async for chunk in cast(AsyncGenerator, completion):
             yield chunk.choices[0].delta.content
@@ -635,13 +635,13 @@ def add_texts_and_embeddings(self, texts: Sequence[Embeddable]) -> None:
         else:
             raise ValueError("Only embeddings of type Text are supported")
         if self._store is None:
-            self._store = self._store_builder(  # type: ignore
+            self._store = self._store_builder(
                 vec_store_text_and_embeddings,
                 texts,
             )
             if self._store is None or not hasattr(self._store, "add_embeddings"):
                 raise ValueError("store_builder did not return a valid vectorstore")
-        self._store.add_embeddings(  # type: ignore
+        self._store.add_embeddings(
             vec_store_text_and_embeddings,
             metadatas=texts,
         )
diff --git a/paperqa/types.py b/paperqa/types.py
index 0a954b5ac..8d051db83 100644
--- a/paperqa/types.py
+++ b/paperqa/types.py
@@ -192,7 +192,7 @@ def remove_computed(cls, data: Any) -> Any:
             data.pop("used_contexts", None)
         return data
 
-    @computed_field  # type: ignore
+    @computed_field  # type: ignore[misc]
     @property
     def used_contexts(self) -> set[str]:
         """Return the used contexts."""
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 000000000..b0ff3d329
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,49 @@
+[tool.mypy]
+# Type-checks the interior of functions without type annotations.
+check_untyped_defs = true
+# Allows enabling one or multiple error codes globally. Note: This option will
+# override disabled error codes from the disable_error_code option.
+enable_error_code = [
+    "ignore-without-code",
+    "mutable-override",
+    "redundant-cast",
+    "redundant-expr",
+    "redundant-self",
+    "truthy-bool",
+    "truthy-iterable",
+    "unreachable",
+    "unused-awaitable",
+    "unused-ignore",
+]
+# Shows a short summary line after error messages.
+error_summary = false
+# Use visually nicer output in error messages: use soft word wrap, show source
+# code snippets, and show error location markers.
+pretty = true
+# Shows column numbers in error messages.
+show_column_numbers = true
+# Shows error codes in error messages.
+# SEE: https://mypy.readthedocs.io/en/stable/error_codes.html#error-codes
+show_error_codes = true
+# Prefixes each error with the relevant context.
+show_error_context = true
+# Warns about casting an expression to its inferred type.
+warn_redundant_casts = true
+# Shows a warning when encountering any code inferred to be unreachable or
+# redundant after performing type analysis.
+warn_unreachable = true
+# Warns about per-module sections in the config file that do not match any
+# files processed when invoking mypy.
+warn_unused_configs = true
+# Warns about unneeded `# type: ignore` comments.
+warn_unused_ignores = true
+
+[[tool.mypy.overrides]]
+# Suppresses error messages about imports that cannot be resolved.
+ignore_missing_imports = true
+# Per-module configuration options
+module = [
+    "fitz",
+    "pyzotero",  # SEE: https://github.com/urschrei/pyzotero/issues/110
+    "sentence_transformers",  # SEE: https://github.com/UKPLab/sentence-transformers/issues/1723
+]
diff --git a/tests/test_paperqa.py b/tests/test_paperqa.py
index 6cea22313..cf2c2a7d6 100644
--- a/tests/test_paperqa.py
+++ b/tests/test_paperqa.py
@@ -413,13 +413,13 @@ async def test_chain_completion(self):
         def accum(x):
             outputs.append(x)
 
-        completion = await call(dict(animal="duck"), callbacks=[accum])
+        completion = await call(dict(animal="duck"), callbacks=[accum])  # type: ignore[call-arg]
         assert completion.seconds_to_first_token > 0
         assert completion.prompt_count > 0
         assert completion.completion_count > 0
         assert str(completion) == "".join(outputs)
 
-        completion = await call(dict(animal="duck"))
+        completion = await call(dict(animal="duck"))  # type: ignore[call-arg]
         assert completion.seconds_to_first_token == 0
         assert completion.seconds_to_last_token > 0
 
@@ -438,13 +438,13 @@ async def test_chain_chat(self):
         def accum(x):
             outputs.append(x)
 
-        completion = await call(dict(animal="duck"), callbacks=[accum])
+        completion = await call(dict(animal="duck"), callbacks=[accum])  # type: ignore[call-arg]
         assert completion.seconds_to_first_token > 0
         assert completion.prompt_count > 0
         assert completion.completion_count > 0
         assert str(completion) == "".join(outputs)
 
-        completion = await call(dict(animal="duck"))
+        completion = await call(dict(animal="duck"))  # type: ignore[call-arg]
         assert completion.seconds_to_first_token == 0
         assert completion.seconds_to_last_token > 0
 
@@ -452,7 +452,7 @@ def accum(x):
         async def ac(x):
             pass
 
-        completion = await call(dict(animal="duck"), callbacks=[accum, ac])
+        completion = await call(dict(animal="duck"), callbacks=[accum, ac])  # type: ignore[call-arg]
 
 
 def test_docs():
@@ -474,7 +474,7 @@ def test_evidence():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     evidence = docs.get_evidence(
         Answer(question="For which state was Bates a governor?"), k=1, max_sources=1
     )
@@ -507,7 +507,7 @@ def test_json_evidence():
         summary_llm_model=summary_llm,
         llm_result_callback=print_callback,
     )
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     evidence = docs.get_evidence(
         Answer(question="For which state was Bates a governor?"), k=1, max_sources=1
     )
@@ -551,7 +551,7 @@ def test_custom_json_props():
         summary_llm_model=summary_llm,
         llm_result_callback=print_callback,
     )
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     evidence = docs.get_evidence(
         Answer(question="For which state was Bates a governor?"), k=1, max_sources=1
     )
@@ -664,7 +664,7 @@ def test_sentence_transformer_embedding():
         citation="WikiMedia Foundation, 2023, Accessed now",
         dockey="test",
     )
-    assert any(docs.docs["test"].embedding)
+    assert any(docs.docs["test"].embedding)  # type: ignore[arg-type]
 
     docs = Docs(
         texts_index=NumpyVectorStore(
@@ -679,7 +679,7 @@ def test_sentence_transformer_embedding():
         citation="WikiMedia Foundation, 2023, Accessed now",
         dockey="test",
     )
-    assert any(docs.docs["test"].embedding)
+    assert any(docs.docs["test"].embedding)  # type: ignore[arg-type]
 
 
 def test_custom_llm():
@@ -762,7 +762,7 @@ def test_langchain_llm():
         get_callbacks=lambda x: [lambda y: print(y, end="")],
     )
 
-    assert docs.summary_llm_model.llm_type == "completion"
+    assert docs.summary_llm_model.llm_type == "completion"  # type: ignore[union-attr]
 
     # trying without callbacks (different codepath)
     docs.get_evidence(
@@ -823,19 +823,19 @@ async def test_langchain_vector_store(self):
         try:
             index = LangchainVectorStore()
             index.add_texts_and_embeddings(some_texts)
-            raise "Failed to check for builder"
+            raise "Failed to check for builder"  # type: ignore[misc]
         except ValueError:
             pass
 
         try:
             index = LangchainVectorStore(store_builder=lambda x: None)
-            raise "Failed to count arguments"
+            raise "Failed to count arguments"  # type: ignore[misc]
         except ValueError:
             pass
 
         try:
             index = LangchainVectorStore(store_builder="foo")
-            raise "Failed to check if builder is callable"
+            raise "Failed to check if builder is callable"  # type: ignore[misc]
         except ValueError:
             pass
 
@@ -847,7 +847,7 @@ async def test_langchain_vector_store(self):
         index.add_texts_and_embeddings(some_texts)
         assert index._store is not None
         # check search returns Text obj
-        data, score = await index.similarity_search(None, "test", k=1)
+        data, score = await index.similarity_search(None, "test", k=1)  # type: ignore[unreachable]
         print(data)
         assert type(data[0]) == Text
 
@@ -938,8 +938,8 @@ def test_docs_pickle():
     )
     assert docs._client is not None
     old_config = docs.llm_model.config
-    old_sconfig = docs.summary_llm_model.config
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now", chunk_chars=1000)
+    old_sconfig = docs.summary_llm_model.config  # type: ignore[union-attr]
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now", chunk_chars=1000)  # type: ignore[arg-type]
     os.remove(doc_path)
     docs_pickle = pickle.dumps(docs)
     docs2 = pickle.loads(docs_pickle)
@@ -989,7 +989,7 @@ def test_bad_context():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     answer = docs.query("What is the radius of Jupyter?")
     assert "cannot answer" in answer.answer
     os.remove(doc_path)
@@ -1004,9 +1004,9 @@ def test_repeat_keys():
     docs = Docs(
         llm_model=OpenAILLMModel(config=dict(temperature=0.0, model="babbage-002"))
     )
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     try:
-        docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+        docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     except ValueError:
         pass
     assert len(docs.docs) == 1
@@ -1017,7 +1017,7 @@ def test_repeat_keys():
         # get wiki page about politician
         f.write(r.text)
         f.write("\n")  # so we don't have same hash
-    docs.add(doc_path2, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path2, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     assert len(docs.docs) == 2
 
     # check keys
@@ -1033,7 +1033,7 @@ def test_pdf_reader():
     tests_dir = os.path.dirname(os.path.abspath(__file__))
     doc_path = os.path.join(tests_dir, "paper.pdf")
     docs = Docs(llm_model=OpenAILLMModel(config=dict(temperature=0.0, model="gpt-4")))
-    docs.add(doc_path, "Wellawatte et al, XAI Review, 2023")
+    docs.add(doc_path, "Wellawatte et al, XAI Review, 2023")  # type: ignore[arg-type]
     answer = docs.query("Are counterfactuals actionable? [yes/no]")
     assert "yes" in answer.answer or "Yes" in answer.answer
 
@@ -1067,14 +1067,14 @@ def test_pdf_pypdf_reader():
     tests_dir = os.path.dirname(os.path.abspath(__file__))
     doc_path = os.path.join(tests_dir, "paper.pdf")
     splits1 = read_doc(
-        doc_path,
+        doc_path,  # type: ignore[arg-type]
         Doc(docname="foo", citation="Foo et al, 2002", dockey="1"),
         force_pypdf=True,
         overlap=100,
         chunk_chars=3000,
     )
     splits2 = read_doc(
-        doc_path,
+        doc_path,  # type: ignore[arg-type]
         Doc(docname="foo", citation="Foo et al, 2002", dockey="1"),
         force_pypdf=False,
         overlap=100,
@@ -1093,7 +1093,7 @@ def test_prompt_length():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     docs.query("What is the name of the politician?")
 
 
@@ -1103,7 +1103,7 @@ def test_code():
     docs = Docs(
         llm_model=OpenAILLMModel(config=dict(temperature=0.0, model="babbage-002"))
     )
-    docs.add(doc_path, "test_paperqa.py", docname="test_paperqa.py", disable_check=True)
+    docs.add(doc_path, "test_paperqa.py", docname="test_paperqa.py", disable_check=True)  # type: ignore[arg-type]
     assert len(docs.docs) == 1
     docs.query("What function tests the preview?")
 
@@ -1115,7 +1115,7 @@ def test_citation():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path)
+    docs.add(doc_path)  # type: ignore[arg-type]
     assert (
         list(docs.docs.values())[0].docname == "Wikipedia2024"
         or list(docs.docs.values())[0].docname == "Frederick2024"
@@ -1132,12 +1132,12 @@ def test_dockey_filter():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     # add with new dockey
     with open("example.txt", "w", encoding="utf-8") as f:
         f.write(r.text)
         f.write("\n")  # so we don't have same hash
-    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", dockey="test")
+    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", dockey="test")  # type: ignore[arg-type]
     answer = Answer(question="What country is Bates from?", dockey_filter=["test"])
     docs.get_evidence(answer)
 
@@ -1150,12 +1150,12 @@ def test_dockey_delete():
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
     docs = Docs()
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     # add with new dockey
     with open("example.txt", "w", encoding="utf-8") as f:
         f.write(r.text)
         f.write("\n\nBates could be from Angola")  # so we don't have same hash
-    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", docname="test")
+    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", docname="test")  # type: ignore[arg-type]
     answer = Answer(question="What country was Bates born in?")
     answer = docs.get_evidence(
         answer, max_sources=25, k=30
@@ -1182,14 +1182,14 @@ def test_query_filter():
         f.write(r.text)
     docs = Docs()
     docs.add(
-        doc_path,
+        doc_path,  # type: ignore[arg-type]
         "Information about Fredrick Bates, WikiMedia Foundation, 2023, Accessed now",
     )
     # add with new dockey
     with open("example.txt", "w", encoding="utf-8") as f:
         f.write(r.text)
         f.write("\n")  # so we don't have same hash
-    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", dockey="test")
+    docs.add("example.txt", "WikiMedia Foundation, 2023, Accessed now", dockey="test")  # type: ignore[arg-type]
     docs.query("What country is Bates from?", key_filter=True)
     # the filter shouldn't trigger, so just checking that it doesn't crash
 
@@ -1212,13 +1212,13 @@ def test_too_much_evidence():
         r = requests.get("https://en.wikipedia.org/wiki/Barack_Obama")
         f.write(r.text)
     docs = Docs(llm="gpt-3.5-turbo", summary_llm="gpt-3.5-turbo")
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     # add with new dockey
     with open("example.txt", "w", encoding="utf-8") as f:
         f.write(r.text)
         f.write("\n")  # so we don't have same hash
     docs.add(
-        "example.txt",
+        "example.txt",  # type: ignore[arg-type]
         "WikiMedia Foundation, 2023, Accessed now",
         dockey="test",
         chunk_chars=4000,
@@ -1240,7 +1240,7 @@ def test_custom_prompts():
         # get wiki page about politician
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     answer = docs.query("What country is Frederick Bates from?")
     assert "United States" in answer.answer
 
@@ -1255,7 +1255,7 @@ def test_pre_prompt():
         # get wiki page about politician
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     docs.query("What country is Bates from?")
 
 
@@ -1275,7 +1275,7 @@ def test_post_prompt():
         # get wiki page about politician
         r = requests.get("https://en.wikipedia.org/wiki/Frederick_Bates_(politician)")
         f.write(r.text)
-    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")
+    docs.add(doc_path, "WikiMedia Foundation, 2023, Accessed now")  # type: ignore[arg-type]
     docs.query("What country is Bates from?")
 
     docs = Docs(