Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.lyzr.ai/llms.txt

Use this file to discover all available pages before exploring further.

When Cognis extracts facts from conversation messages, each memory is automatically assigned a category. This enables filtered retrieval and helps organize what the agent knows about a user.

​
Default Categories

CategoryDescriptionExample
identityPersonal identity, name, demographics, age, location”Alice is 28 years old”
relationshipsFamily, friends, social connections, pets”Alice has a dog named Max”
work_careerJob, profession, workplace, colleagues, business”Alice works at Google as a data scientist”
learningEducation, skills, knowledge, certifications, languages”Alice speaks French and Python”
wellnessHealth, fitness, medical conditions, diet, exercise”Alice runs 5K every morning”
lifestyleDaily habits, routines, sleep, transportation”Alice commutes by bike”
interestsHobbies, passions, entertainment, sports, games”Alice loves hiking and Taylor Swift”
preferencesLikes, dislikes, choices, favorites, style”Alice prefers dark mode”
plans_goalsFuture plans, aspirations, goals, dreams, intentions”Alice plans to visit Japan next year”
experiencesPast events, travel, memories, experiences”Alice went to Bali last summer”
opinionsViews, beliefs, attitudes, philosophical”Alice thinks remote work is more productive”
contextSession-specific context, current tasks, immediate needs”Alice is debugging a FastAPI endpoint”
miscAnything that doesn’t fit other categoriesGeneral facts

​
Accessing Categories

Categories are stored in the memory’s metadata: 
resp = m.get_all()
for mem in resp["memories"]:
    print(f"[{mem['metadata']['category']}] {mem['content']}")

# [identity] Alice's name is Alice
# [work_career] Alice works at Google as a data scientist
# [interests] Alice loves hiking

​
Custom Categories

Override the defaults by passing a custom dictionary to CognisConfig: 
from cognis import Cognis, CognisConfig

config = CognisConfig(
    categories={
        "product_feedback": "User feedback about products and features",
        "support_issues": "Technical issues and bug reports",
        "account_info": "Account details, subscription, billing",
        "preferences": "User preferences and settings",
        "misc": "Anything else",
    }
)

m = Cognis(config=config, owner_id="user_1")
The LLM extraction prompt adapts to use your custom categories for classification.

​
How Categorization Works

During add(), the LLM extraction step:
  1. Reads the conversation messages
  2. Extracts discrete facts as individual memory records
  3. Assigns each fact to the most appropriate category from the configured list
  4. Checks for duplicates against existing memories (threshold: 0.85 similarity)
  5. Stores new facts or updates existing ones with versioning