Initial pass at backend changes to support agents

- Add a db model for Agents, attaching them to conversations
- When an agent is added to a conversation, override the system prompt to tweak the instructions
- Agents can be configured with prompt modification, model specification, a profile picture, and other things
- Admin-configured models will not be editable by individual users
- Add unit tests to verify agent behavior. Unit tests demonstrate imperfect adherence to prompt specifications
This commit is contained in:
sabaimran
2024-03-11 12:45:24 +05:30
parent 1da453306e
commit 9b88976f36
14 changed files with 428 additions and 58 deletions

View File

@@ -16,6 +16,7 @@ from pgvector.django import CosineDistance
from torch import Tensor
from khoj.database.models import (
Agent,
ChatModelOptions,
ClientApplication,
Conversation,
@@ -391,6 +392,12 @@ class ClientApplicationAdapters:
return await ClientApplication.objects.filter(client_id=client_id, client_secret=client_secret).afirst()
class AgentAdapters:
@staticmethod
async def aget_agent_by_id(agent_id: int):
return await Agent.objects.filter(id=agent_id).afirst()
class ConversationAdapters:
@staticmethod
def get_conversation_by_user(
@@ -431,7 +438,12 @@ class ConversationAdapters:
return Conversation.objects.filter(id=conversation_id).first()
@staticmethod
async def acreate_conversation_session(user: KhojUser, client_application: ClientApplication = None):
async def acreate_conversation_session(
user: KhojUser, client_application: ClientApplication = None, agent_id: int = None
):
if agent_id:
agent = await AgentAdapters.aget_agent_by_id(id)
return await Conversation.objects.acreate(user=user, client=client_application, agent=agent)
return await Conversation.objects.acreate(user=user, client=client_application)
@staticmethod
@@ -446,7 +458,7 @@ class ConversationAdapters:
conversation = Conversation.objects.filter(user=user, client=client_application).order_by("-updated_at")
if await conversation.aexists():
return await conversation.afirst()
return await conversation.prefetch_related("agent").afirst()
return await Conversation.objects.acreate(user=user, client=client_application, slug=slug)
@@ -606,9 +618,14 @@ class ConversationAdapters:
return random.sample(all_questions, max_results)
@staticmethod
def get_valid_conversation_config(user: KhojUser):
def get_valid_conversation_config(user: KhojUser, conversation: Conversation):
offline_chat_config = ConversationAdapters.get_offline_chat_conversation_config()
conversation_config = ConversationAdapters.get_conversation_config(user)
if conversation.agent and conversation.agent.chat_model:
conversation_config = conversation.agent.chat_model
else:
conversation_config = ConversationAdapters.get_conversation_config(user)
if conversation_config is None:
conversation_config = ConversationAdapters.get_default_conversation_config()

View File

@@ -6,6 +6,7 @@ from django.contrib.auth.admin import UserAdmin
from django.http import HttpResponse
from khoj.database.models import (
Agent,
ChatModelOptions,
ClientApplication,
Conversation,
@@ -50,6 +51,7 @@ admin.site.register(ReflectiveQuestion)
admin.site.register(UserSearchModelConfig)
admin.site.register(TextToImageModelConfig)
admin.site.register(ClientApplication)
admin.site.register(Agent)
@admin.register(Entry)

View File

@@ -0,0 +1,52 @@
# Generated by Django 4.2.10 on 2024-03-11 05:12
import django.db.models.deletion
from django.conf import settings
from django.db import migrations, models
class Migration(migrations.Migration):
dependencies = [
("database", "0030_conversation_slug_and_title"),
]
operations = [
migrations.CreateModel(
name="Agent",
fields=[
("id", models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name="ID")),
("created_at", models.DateTimeField(auto_now_add=True)),
("updated_at", models.DateTimeField(auto_now=True)),
("name", models.CharField(max_length=200)),
("tuning", models.TextField()),
("avatar", models.URLField(blank=True, default=None, max_length=400, null=True)),
("tools", models.JSONField(default=list)),
("public", models.BooleanField(default=False)),
("managed_by_admin", models.BooleanField(default=False)),
(
"chat_model",
models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to="database.chatmodeloptions"),
),
(
"creator",
models.ForeignKey(
blank=True,
default=None,
null=True,
on_delete=django.db.models.deletion.CASCADE,
to=settings.AUTH_USER_MODEL,
),
),
],
options={
"abstract": False,
},
),
migrations.AddField(
model_name="conversation",
name="agent",
field=models.ForeignKey(
blank=True, default=None, null=True, on_delete=django.db.models.deletion.SET_NULL, to="database.agent"
),
),
]

View File

@@ -69,6 +69,28 @@ class Subscription(BaseModel):
renewal_date = models.DateTimeField(null=True, default=None, blank=True)
class ChatModelOptions(BaseModel):
class ModelType(models.TextChoices):
OPENAI = "openai"
OFFLINE = "offline"
max_prompt_size = models.IntegerField(default=None, null=True, blank=True)
tokenizer = models.CharField(max_length=200, default=None, null=True, blank=True)
chat_model = models.CharField(max_length=200, default="mistral-7b-instruct-v0.1.Q4_0.gguf")
model_type = models.CharField(max_length=200, choices=ModelType.choices, default=ModelType.OFFLINE)
class Agent(BaseModel):
creator = models.ForeignKey(KhojUser, on_delete=models.CASCADE, default=None, null=True, blank=True)
name = models.CharField(max_length=200)
tuning = models.TextField()
avatar = models.URLField(max_length=400, default=None, null=True, blank=True)
tools = models.JSONField(default=list) # List of tools the agent has access to, like online search or notes search
public = models.BooleanField(default=False)
managed_by_admin = models.BooleanField(default=False)
chat_model = models.ForeignKey(ChatModelOptions, on_delete=models.CASCADE)
class NotionConfig(BaseModel):
token = models.CharField(max_length=200)
user = models.ForeignKey(KhojUser, on_delete=models.CASCADE)
@@ -153,17 +175,6 @@ class SpeechToTextModelOptions(BaseModel):
model_type = models.CharField(max_length=200, choices=ModelType.choices, default=ModelType.OFFLINE)
class ChatModelOptions(BaseModel):
class ModelType(models.TextChoices):
OPENAI = "openai"
OFFLINE = "offline"
max_prompt_size = models.IntegerField(default=None, null=True, blank=True)
tokenizer = models.CharField(max_length=200, default=None, null=True, blank=True)
chat_model = models.CharField(max_length=200, default="mistral-7b-instruct-v0.1.Q4_0.gguf")
model_type = models.CharField(max_length=200, choices=ModelType.choices, default=ModelType.OFFLINE)
class UserConversationConfig(BaseModel):
user = models.OneToOneField(KhojUser, on_delete=models.CASCADE)
setting = models.ForeignKey(ChatModelOptions, on_delete=models.CASCADE, default=None, null=True, blank=True)
@@ -180,6 +191,7 @@ class Conversation(BaseModel):
client = models.ForeignKey(ClientApplication, on_delete=models.CASCADE, default=None, null=True, blank=True)
slug = models.CharField(max_length=200, default=None, null=True, blank=True)
title = models.CharField(max_length=200, default=None, null=True, blank=True)
agent = models.ForeignKey(Agent, on_delete=models.SET_NULL, default=None, null=True, blank=True)
class ReflectiveQuestion(BaseModel):

View File

@@ -6,6 +6,7 @@ from typing import Any, Iterator, List, Union
from langchain.schema import ChatMessage
from khoj.database.models import Agent
from khoj.processor.conversation import prompts
from khoj.processor.conversation.utils import (
ThreadedGenerator,
@@ -141,6 +142,7 @@ def converse_offline(
tokenizer_name=None,
location_data: LocationData = None,
user_name: str = None,
agent: Agent = None,
) -> Union[ThreadedGenerator, Iterator[str]]:
"""
Converse with user using Llama
@@ -156,6 +158,16 @@ def converse_offline(
# Initialize Variables
compiled_references_message = "\n\n".join({f"{item}" for item in references})
system_prompt = ""
current_date = datetime.now().strftime("%Y-%m-%d")
if agent:
system_prompt = prompts.custom_system_prompt_message_gpt4all.format(
name=agent.name, bio=agent.tuning, current_date=current_date
)
else:
system_prompt = prompts.system_prompt_message_gpt4all.format(current_date=current_date)
conversation_primer = prompts.query_prompt.format(query=user_query)
if location_data:
@@ -185,10 +197,9 @@ def converse_offline(
conversation_primer = f"{prompts.notes_conversation_gpt4all.format(references=compiled_references_message)}\n{conversation_primer}"
# Setup Prompt with Primer or Conversation History
current_date = datetime.now().strftime("%Y-%m-%d")
messages = generate_chatml_messages_with_context(
conversation_primer,
prompts.system_prompt_message_gpt4all.format(current_date=current_date),
system_prompt,
conversation_log,
model_name=model,
max_prompt_size=max_prompt_size,

View File

@@ -5,6 +5,7 @@ from typing import Optional
from langchain.schema import ChatMessage
from khoj.database.models import Agent
from khoj.processor.conversation import prompts
from khoj.processor.conversation.openai.utils import (
chat_completion_with_backoff,
@@ -118,6 +119,7 @@ def converse(
tokenizer_name=None,
location_data: LocationData = None,
user_name: str = None,
agent: Agent = None,
):
"""
Converse with user using OpenAI's ChatGPT
@@ -128,6 +130,13 @@ def converse(
conversation_primer = prompts.query_prompt.format(query=user_query)
system_prompt = ""
if agent:
system_prompt = prompts.custom_personality.format(name=agent.name, bio=agent.tuning, current_date=current_date)
else:
system_prompt = prompts.personality.format(current_date=current_date)
if location_data:
location = f"{location_data.city}, {location_data.region}, {location_data.country}"
location_prompt = prompts.user_location.format(location=location)
@@ -158,7 +167,7 @@ def converse(
# Setup Prompt with Primer or Conversation History
messages = generate_chatml_messages_with_context(
conversation_primer,
prompts.personality.format(current_date=current_date),
system_prompt,
conversation_log,
model,
max_prompt_size,

View File

@@ -21,6 +21,24 @@ Today is {current_date} in UTC.
""".strip()
)
custom_personality = PromptTemplate.from_template(
"""
Your are {name}, a personal agent on Khoj.
Use your general knowledge and past conversation with the user as context to inform your responses.
You were created by Khoj Inc. with the following capabilities:
- You *CAN REMEMBER ALL NOTES and PERSONAL INFORMATION FOREVER* that the user ever shares with you.
- Users can share files and other information with you using the Khoj Desktop, Obsidian or Emacs app. They can also drag and drop their files into the chat window.
- Say "I don't know" or "I don't understand" if you don't know what to say or if you don't know the answer to a question.
- Ask crisp follow-up questions to get additional context, when the answer cannot be inferred from the provided notes or past conversations.
- Sometimes the user will share personal information that needs to be remembered, like an account ID or a residential address. These can be acknowledged with a simple "Got it" or "Okay".
Today is {current_date} in UTC.
Here's a bio about you: {bio}
""".strip()
)
## General Conversation
## --
general_conversation = PromptTemplate.from_template(
@@ -61,6 +79,20 @@ Today is {current_date} in UTC.
""".strip()
)
custom_system_prompt_message_gpt4all = PromptTemplate.from_template(
"""
You are {name}, a personal agent on Khoj.
- Use your general knowledge and past conversation with the user as context to inform your responses.
- If you do not know the answer, say 'I don't know.'
- Think step-by-step and ask questions to get the necessary information to answer the user's question.
- Do not print verbatim Notes unless necessary.
Today is {current_date} in UTC.
Here is your instruction set:\n{bio}
""".strip()
)
system_prompt_message_extract_questions_gpt4all = f"""You are Khoj, a kind and intelligent personal assistant. When the user asks you a question, you ask follow-up questions to clarify the necessary information you need in order to answer from the user's perspective.
- Write the question as if you can search for the answer on the user's personal notes.
- Try to be as specific as possible. Instead of saying "they" or "it" or "he", use the name of the person or thing you are referring to. For example, instead of saying "Which store did they go to?", say "Which store did Alice and Bob go to?".

View File

@@ -148,6 +148,7 @@ def chat_sessions(
async def create_chat_session(
request: Request,
common: CommonQueryParams,
agent_id: Optional[int] = None,
):
user = request.user.object
@@ -250,9 +251,11 @@ async def chat(
formatted_help = help_message.format(model=model_type, version=state.khoj_version, device=get_device())
return StreamingResponse(iter([formatted_help]), media_type="text/event-stream", status_code=200)
meta_log = (
await ConversationAdapters.aget_conversation_by_user(user, request.user.client_app, conversation_id, slug)
).conversation_log
conversation = await ConversationAdapters.aget_conversation_by_user(
user, request.user.client_app, conversation_id, slug
)
meta_log = conversation.conversation_log
if conversation_commands == [ConversationCommand.Default]:
conversation_commands = await aget_relevant_information_sources(q, meta_log)
@@ -335,6 +338,7 @@ async def chat(
llm_response, chat_metadata = await agenerate_chat_response(
defiltered_query,
meta_log,
conversation,
compiled_references,
online_results,
inferred_queries,

View File

@@ -12,8 +12,10 @@ from starlette.authentication import has_required_scope
from khoj.database.adapters import ConversationAdapters, EntryAdapters
from khoj.database.models import (
Agent,
ChatModelOptions,
ClientApplication,
Conversation,
KhojUser,
Subscription,
TextToImageModelConfig,
@@ -359,6 +361,7 @@ async def send_message_to_model_wrapper(
def generate_chat_response(
q: str,
meta_log: dict,
conversation: Conversation,
compiled_references: List[str] = [],
online_results: Dict[str, Any] = {},
inferred_queries: List[str] = [],
@@ -374,6 +377,7 @@ def generate_chat_response(
logger.debug(f"Conversation Types: {conversation_commands}")
metadata = {}
agent = conversation.agent
try:
partial_completion = partial(
@@ -388,7 +392,7 @@ def generate_chat_response(
conversation_id=conversation_id,
)
conversation_config = ConversationAdapters.get_valid_conversation_config(user)
conversation_config = ConversationAdapters.get_valid_conversation_config(user, conversation)
if conversation_config.model_type == "offline":
if state.gpt4all_processor_config is None or state.gpt4all_processor_config.loaded_model is None:
state.gpt4all_processor_config = GPT4AllProcessorModel(conversation_config.chat_model)
@@ -407,6 +411,7 @@ def generate_chat_response(
tokenizer_name=conversation_config.tokenizer,
location_data=location_data,
user_name=user_name,
agent=agent,
)
elif conversation_config.model_type == "openai":
@@ -426,6 +431,7 @@ def generate_chat_response(
tokenizer_name=conversation_config.tokenizer,
location_data=location_data,
user_name=user_name,
agent=agent,
)
metadata.update({"chat_model": conversation_config.chat_model})