Implementing Long-Term Memory using Julep

Talk to Jessica, a stuck-up Cali teen who doesn't forget.

Julep is an open-source platform for AI applications. It offers long-term memories, context management, and supports multi-user and multi-agent apps.

We're going to write a small conversational bot which has the ability to form long-term memories using Julep.

Join our Discord server to see it in action!

!pip install julep -U --quiet

Run to view results

from julep import Client from pprint import pprint import textwrap import os base_url = os.environ.get("JULEP_API_URL") api_key = os.environ.get("JULEP_API_KEY") client = Client(api_key=api_key, base_url=base_url)

Run to view results

Creating an "agent"

Agent is the object to which LLM settings, like model, temperature along with tools are scoped to.

gif_tool = { "type": "function", "function": { "name": "get_gif", "description": "Retrieves a relevant GIF for a given search string. The string must reflect the context of the conversation. The tool swill return the appropriate GIF for the search keyword. It should be used whenever deemed more appropriate then replying. If message can be conveyed in a GIF, then it shall be used. Or if the user specifically asks for the response using a GIF", "parameters": { "type": "object", "properties": { "keywords": { "type": "string", "description": "Search keyword for the GIF, e.g. 'that friday feeling', 'Eat Sh*t'", } }, "required": ["keywords"], }, }, } name = "Jessica" about = "Jessica is a stuck up Cali teenager. Showing rebellion is an evolutionary necessity for her." default_settings = { "temperature": 0.7, "top_p": 1, "min_p": 0.01, "presence_penalty": 0, "frequency_penalty": 0, "length_penalty": 1.0, "max_tokens": 150, } agent = client.agents.create( name=name, about=about, default_settings=default_settings, model="gpt-4", # tools=[gif_tool], )

Run to view results

Creating a "user"

User is the object which represents the user of the application.

Memories are formed and saved for each user and many users can talk to one agent.

about = """ Average nerdy techbro/girl who spends 8 hours a day in front of a laptop. Thinks they can build a small SaaS tool and gain financial independence within the year. """ user = client.users.create( name="Anon", about=about, )

Run to view results

Putting it all together in a "session"

A "user" and an "agent" communicate in a "session". System prompt goes here.

Conversation history and summary are stored in a "session" which saves the conversation history.

The session paradigm allows for; many users to interact with one agent and allow separation of conversation history and memories.

situation_prompt = """You are Jessica. You're a stuck up Cali teenager. You basically complain about everything. You live in Bel-Air, Los Angeles and drag yourself to Curtis High School when you must. You don't like basic broke ppl like those who own an Android or don't bring their Mercedes to school. You're surrounded by annoying 'well-meaning' people who don't understand your vibe. You're simply better and classier than the dumbasses you are surrounded by. Recent Tweets 1. 'my step mom is so fake and nice all the time? she's not even rich wtf. why tf did my dad marry her.' 2. 'am i the only one who thinks jordan belfort was a good guy??? he's so cute like..'' 3. 'if your butler doesn't know the difference soda and Perrier don't even fkng talk to me' Here, you are hanging out in a Discord Server called Julep AI to troll these nerdy losers. """ session = client.sessions.create( user_id=user.id, agent_id=agent.id, situation=situation_prompt )

Run to view results

user_msg

response = client.sessions.chat( session_id=session.id, messages=[ { "role": "user", "content": user_msg, "name": "Anon", } ], recall=True, remember=True, ) print("\n".join(textwrap.wrap(response.response[0][0].content, width=100)))

Run to view results

print(client.sessions.history(session_id=session.id))

Run to view results

.css-15w88e5{color:var(--chakra-colors-fg-neutral-primary);font-weight:inherit;letter-spacing:-0.09px;}Implementing Long-Term Memory using Julep

Creating an "agent"

Creating a "user"

Putting it all together in a "session"

Implementing Long-Term Memory using Julep