support chat history (#438)

Co-authored-by: Philip Meier <[email protected]>

docs/examples/gallery_streaming.py

@@ -44,8 +44,8 @@
 
 
 class DemoStreamingAssistant(assistants.RagnaDemoAssistant):
-    def answer(self, prompt, sources):
-        content = next(super().answer(prompt, sources))
+    def answer(self, messages):
+        content = next(super().answer(messages))
         for chunk in content.split(" "):
             yield f"{chunk} "
 

docs/tutorials/gallery_custom_components.py

@@ -30,7 +30,7 @@
 
 import uuid
 
-from ragna.core import Document, Source, SourceStorage
+from ragna.core import Document, Source, SourceStorage, Message
 
 
 class TutorialSourceStorage(SourceStorage):
@@ -61,9 +61,9 @@ def retrieve(
 # %%
 # ### Assistant
 #
-# [ragna.core.Assistant][]s are objects that take a user prompt and relevant
-# [ragna.core.Source][]s and generate a response form that. Usually, assistants are
-# LLMs.
+# [ragna.core.Assistant][]s are objects that take the chat history as list of
+# [ragna.core.Message][]s and their relevant [ragna.core.Source][]s and generate a
+# response from that. Usually, assistants are LLMs.
 #
 # In this tutorial, we define a minimal `TutorialAssistant` that is similar to
 # [ragna.assistants.RagnaDemoAssistant][]. In `.answer()` we mirror back the user
@@ -82,8 +82,11 @@ def retrieve(
 
 
 class TutorialAssistant(Assistant):
-    def answer(self, prompt: str, sources: list[Source]) -> Iterator[str]:
+    def answer(self, messages: list[Message]) -> Iterator[str]:
         print(f"Running {type(self).__name__}().answer()")
+        # For simplicity, we only deal with the last message here, i.e. the latest user
+        # prompt.
+        prompt, sources = (message := messages[-1]).content, message.sources
         yield (
             f"To answer the user prompt '{prompt}', "
             f"I was given {len(sources)} source(s)."
@@ -254,8 +257,7 @@ def answer(self, prompt: str, sources: list[Source]) -> Iterator[str]:
 class ElaborateTutorialAssistant(Assistant):
     def answer(
         self,
-        prompt: str,
-        sources: list[Source],
+        messages: list[Message],
         *,
         my_required_parameter: int,
         my_optional_parameter: str = "foo",
@@ -393,9 +395,7 @@ def answer(
 
 
 class AsyncAssistant(Assistant):
-    async def answer(
-        self, prompt: str, sources: list[Source]
-    ) -> AsyncIterator[str]:
+    async def answer(self, messages: list[Message]) -> AsyncIterator[str]:
         print(f"Running {type(self).__name__}().answer()")
         start = time.perf_counter()
         await asyncio.sleep(0.3)

ragna/assistants/_ai21labs.py

@@ -1,6 +1,6 @@
 from typing import AsyncIterator, cast
 
-from ragna.core import Source
+from ragna.core import Message, Source
 
 from ._http_api import HttpApiAssistant
 
@@ -23,11 +23,12 @@ def _make_system_content(self, sources: list[Source]) -> str:
         return instruction + "\n\n".join(source.content for source in sources)
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
         # See https://docs.ai21.com/reference/j2-chat-api#chat-api-parameters
         # See https://docs.ai21.com/reference/j2-complete-api-ref#api-parameters
         # See https://docs.ai21.com/reference/j2-chat-api#understanding-the-response
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for data in self._call_api(
             "POST",
             f"https://api.ai21.com/studio/v1/j2-{self._MODEL_TYPE}/chat",

ragna/assistants/_anthropic.py

@@ -1,6 +1,6 @@
 from typing import AsyncIterator, cast
 
-from ragna.core import PackageRequirement, RagnaException, Requirement, Source
+from ragna.core import Message, PackageRequirement, RagnaException, Requirement, Source
 
 from ._http_api import HttpApiAssistant, HttpStreamingProtocol
 
@@ -37,10 +37,11 @@ def _instructize_system_prompt(self, sources: list[Source]) -> str:
         )
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
         # See https://docs.anthropic.com/claude/reference/messages_post
         # See https://docs.anthropic.com/claude/reference/streaming
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for data in self._call_api(
             "POST",
             "https://api.anthropic.com/v1/messages",

ragna/assistants/_cohere.py

@@ -1,6 +1,6 @@
 from typing import AsyncIterator, cast
 
-from ragna.core import RagnaException, Source
+from ragna.core import Message, RagnaException, Source
 
 from ._http_api import HttpApiAssistant, HttpStreamingProtocol
 
@@ -25,11 +25,12 @@ def _make_source_documents(self, sources: list[Source]) -> list[dict[str, str]]:
         return [{"title": source.id, "snippet": source.content} for source in sources]
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
         # See https://docs.cohere.com/docs/cochat-beta
         # See https://docs.cohere.com/reference/chat
         # See https://docs.cohere.com/docs/retrieval-augmented-generation-rag
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for event in self._call_api(
             "POST",
             "https://api.cohere.ai/v1/chat",

ragna/assistants/_demo.py

@@ -1,8 +1,7 @@
-import re
 import textwrap
 from typing import Iterator
 
-from ragna.core import Assistant, Source
+from ragna.core import Assistant, Message, MessageRole
 
 
 class RagnaDemoAssistant(Assistant):
@@ -22,11 +21,11 @@ class RagnaDemoAssistant(Assistant):
     def display_name(cls) -> str:
         return "Ragna/DemoAssistant"
 
-    def answer(self, prompt: str, sources: list[Source]) -> Iterator[str]:
-        if re.search("markdown", prompt, re.IGNORECASE):
+    def answer(self, messages: list[Message]) -> Iterator[str]:
+        if "markdown" in messages[-1].content.lower():
             yield self._markdown_answer()
         else:
-            yield self._default_answer(prompt, sources)
+            yield self._default_answer(messages)
 
     def _markdown_answer(self) -> str:
         return textwrap.dedent(
@@ -39,7 +38,8 @@ def _markdown_answer(self) -> str:
             """
         ).strip()
 
-    def _default_answer(self, prompt: str, sources: list[Source]) -> str:
+    def _default_answer(self, messages: list[Message]) -> str:
+        prompt, sources = (message := messages[-1]).content, message.sources
         sources_display = []
         for source in sources:
             source_display = f"- {source.document.name}"
@@ -50,13 +50,16 @@ def _default_answer(self, prompt: str, sources: list[Source]) -> str:
         if len(sources) > 3:
             sources_display.append("[...]")
 
+        n_messages = len([m for m in messages if m.role == MessageRole.USER])
         return (
             textwrap.dedent(
                 """
-                I'm a demo assistant and can be used to try Ragnas workflow.
+                I'm a demo assistant and can be used to try Ragna's workflow.
                 I will only mirror back my inputs. 
+
+                So far I have received {n_messages} messages.
                 
-                Your prompt was:
+                Your last prompt was:
 
                 > {prompt}
 
@@ -66,5 +69,10 @@ def _default_answer(self, prompt: str, sources: list[Source]) -> str:
                 """
             )
             .strip()
-            .format(name=str(self), prompt=prompt, sources="\n".join(sources_display))
+            .format(
+                name=str(self),
+                n_messages=n_messages,
+                prompt=prompt,
+                sources="\n".join(sources_display),
+            )
         )

ragna/assistants/_google.py

@@ -1,6 +1,6 @@
 from typing import AsyncIterator
 
-from ragna.core import Source
+from ragna.core import Message, Source
 
 from ._http_api import HttpApiAssistant, HttpStreamingProtocol
 
@@ -26,8 +26,9 @@ def _instructize_prompt(self, prompt: str, sources: list[Source]) -> str:
         )
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for chunk in self._call_api(
             "POST",
             f"https://generativelanguage.googleapis.com/v1beta/models/{self._MODEL}:streamGenerateContent",

ragna/assistants/_ollama.py

@@ -2,7 +2,7 @@
 from functools import cached_property
 from typing import AsyncIterator, cast
 
-from ragna.core import RagnaException, Source
+from ragna.core import Message, RagnaException
 
 from ._http_api import HttpStreamingProtocol
 from ._openai import OpenaiLikeHttpApiAssistant
@@ -30,8 +30,9 @@ def _url(self) -> str:
         return f"{base_url}/api/chat"
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for data in self._stream(prompt, sources, max_new_tokens=max_new_tokens):
             # Modeled after
             # https://github.com/ollama/ollama/blob/06a1508bfe456e82ba053ea554264e140c5057b5/examples/python-loganalysis/readme.md?plain=1#L57-L62

ragna/assistants/_openai.py

@@ -2,7 +2,7 @@
 from functools import cached_property
 from typing import Any, AsyncIterator, Optional, cast
 
-from ragna.core import Source
+from ragna.core import Message, Source
 
 from ._http_api import HttpApiAssistant, HttpStreamingProtocol
 
@@ -55,8 +55,9 @@ def _stream(
         return self._call_api("POST", self._url, headers=headers, json=json_)
 
     async def answer(
-        self, prompt: str, sources: list[Source], *, max_new_tokens: int = 256
+        self, messages: list[Message], *, max_new_tokens: int = 256
     ) -> AsyncIterator[str]:
+        prompt, sources = (message := messages[-1]).content, message.sources
         async for data in self._stream(prompt, sources, max_new_tokens=max_new_tokens):
             choice = data["choices"][0]
             if choice["finish_reason"] is not None:

ragna/core/_components.py

@@ -147,7 +147,7 @@ def retrieve(self, documents: list[Document], prompt: str) -> list[Source]:
         ...
 
 
-class MessageRole(enum.Enum):
+class MessageRole(str, enum.Enum):
     """Message role
 
     Attributes:
@@ -238,12 +238,12 @@ class Assistant(Component, abc.ABC):
     __ragna_protocol_methods__ = ["answer"]
 
     @abc.abstractmethod
-    def answer(self, prompt: str, sources: list[Source]) -> Iterator[str]:
-        """Answer a prompt given some sources.
+    def answer(self, messages: list[Message]) -> Iterator[str]:
+        """Answer a prompt given the chat history.
 
         Args:
-            prompt: Prompt to be answered.
-            sources: Sources to use when answering answer the prompt.
+            messages: List of messages in the chat history. The last item is the current
+                user prompt and has the relevant sources attached to it.
 
         Returns:
             Answer.

ragna/core/_rag.py

@@ -220,12 +220,13 @@ async def answer(self, prompt: str, *, stream: bool = False) -> Message:
                 detail=RagnaException.EVENT,
             )
 
-        self._messages.append(Message(content=prompt, role=MessageRole.USER))
-
         sources = await self._run(self.source_storage.retrieve, self.documents, prompt)
 
+        question = Message(content=prompt, role=MessageRole.USER, sources=sources)
+        self._messages.append(question)
+
         answer = Message(
-            content=self._run_gen(self.assistant.answer, prompt, sources),
+            content=self._run_gen(self.assistant.answer, self._messages.copy()),
             role=MessageRole.ASSISTANT,
             sources=sources,
         )

ragna/deploy/_api/core.py

@@ -296,9 +296,6 @@ async def answer(
     ) -> schemas.Message:
         with get_session() as session:
             chat = database.get_chat(session, user=user, id=id)
-            chat.messages.append(
-                schemas.Message(content=prompt, role=ragna.core.MessageRole.USER)
-            )
             core_chat = schema_to_core_chat(session, user=user, chat=chat)
 
         # smoke test to catch errors unrelated to streaming
@@ -311,6 +308,13 @@ async def answer(
                 detail=str(e),
             ) from e
 
+        sources = [schemas.Source.from_core(source) for source in core_answer.sources]
+        chat.messages.append(
+            schemas.Message(
+                content=prompt, role=ragna.core.MessageRole.USER, sources=sources
+            )
+        )
+
         if stream:
             # smoke test to catch errors unrelated to streaming
             # longer-term solution tracked by #409
@@ -320,10 +324,7 @@ async def answer(
                 answer = schemas.Message(
                     content=content_chunk,
                     role=core_answer.role,
-                    sources=[
-                        schemas.Source.from_core(source)
-                        for source in core_answer.sources
-                    ],
+                    sources=sources,
                 )
             except Exception as e:
                 raise HTTPException(

tests/assistants/test_api.py

@@ -5,7 +5,7 @@
 from ragna import assistants
 from ragna._compat import anext
 from ragna.assistants._http_api import HttpApiAssistant
-from ragna.core import RagnaException
+from ragna.core import Message, RagnaException
 from tests.utils import skip_on_windows
 
 HTTP_API_ASSISTANTS = [
@@ -25,7 +25,8 @@
 async def test_api_call_error_smoke(mocker, assistant):
     mocker.patch.dict(os.environ, {assistant._API_KEY_ENV_VAR: "SENTINEL"})
 
-    chunks = assistant().answer(prompt="?", sources=[])
+    messages = [Message(content="?", sources=[])]
+    chunks = assistant().answer(messages)
 
     with pytest.raises(RagnaException, match="API call failed"):
         await anext(chunks)

tests/core/test_rag.py

@@ -45,8 +45,7 @@ def test_params_validation_missing(self, demo_document):
         class ValidationAssistant(Assistant):
             def answer(
                 self,
-                prompt,
-                sources,
+                messages,
                 bool_param: bool,
                 int_param: int,
                 float_param: float,
@@ -65,8 +64,7 @@ def test_params_validation_wrong_type(self, demo_document):
         class ValidationAssistant(Assistant):
             def answer(
                 self,
-                prompt,
-                sources,
+                messages,
                 bool_param: bool,
                 int_param: int,
                 float_param: float,