For each user, Memobase will track and model the profile. You can use them in your AI prompt to offer a global understanding of the user.

In practice, the user profile won’t be too large, but it’s better for you to be able to control the final size of you prompt.

Memobase offers some params to help you filter the profiles

​
Filter by Rules

You can pass some rules to Memobase API to filter some profiles:

  • max_token_size: The maximum token size of the profile.
  • prefer_topics: those topics will be ranked higher, so they are more likely to be included in the final profile.
  • only_topics: only include these topics, other topics will be filtered.

Detailed params can be found in API docs.

​
Filter by Context

Memobase also offers a semantic filter to filter the profiles. You can pass the latest chat messages to Memobase, and get the “contextual” profiles. “Contextual” here is much more better than a embedding-based search: Memobase uses LLM to reason and judge which profile may help the future chatting.

So if user just say ""Find some restaurants for me”, Memobase will filter contact_info::city, interest::foods, health::allergies and so on. Those profiles will be randed higher in the results.

from memobase import Memobase

client = Memobase(project_url='PROJECT_URL', api_key='PROJECT_TOKEN')
uid = client.add_user()
u = client.get_user(uid)

p = u.profile(chats = [
  {"role": "user", "content": "Find some restaurants for me"},
  need_json=True
])

print(p)