kemi 0.1.7

Persistent memory for AI agents. Three methods. Zero infra.

kemi

Persistent memory for AI agents. Three methods. Zero infra.

from kemi import Memory

memory = Memory()  # SQLite + local embeddings, no API keys needed

memory.remember("user123", "User prefers dark mode")
memory.remember("user123", "User is vegetarian")

results = memory.recall("user123", "what are the user's preferences?")
# Returns ranked, deduplicated memories

memory.forget("user123")  # GDPR-compliant deletion

Install

# Zero dependencies — SQLite storage, no embedding (bring your own)
pip install kemi

# With local embeddings (no API key needed, ~130MB model download)
pip install kemi[local]

# With OpenAI embeddings
pip install kemi[openai]

Why kemi

Every existing memory library either hosts your data on their servers, requires Docker and 4 services to run, or locks you into a specific framework.

kemi is different:

• Zero infrastructure — runs on your machine, single pip install
• Your data — never leaves your machine, stored in SQLite by default
• Bring your own embedding — OpenAI, local models, or any function
• Framework agnostic — works with LangChain, CrewAI, AutoGen, or plain Python
• 100% free — MIT license, no paid tiers, no cloud lock-in

Usage

Zero-config (local embeddings)

from kemi import Memory

memory = Memory()
memory.remember("user123", "User is vegetarian", importance=0.9)
results = memory.recall("user123", "food preferences")

With OpenAI embeddings

from kemi import Memory
from kemi.adapters.embedding.openai import OpenAIEmbedAdapter

memory = Memory(embed=OpenAIEmbedAdapter())
memory.remember("user123", "User prefers concise responses")
results = memory.recall("user123", "communication style")

Async usage (FastAPI, asyncio)

from fastapi import FastAPI
from kemi import Memory

app = FastAPI()
memory = Memory()

@app.post("/chat")
async def chat(user_id: str, message: str):
    await memory.aremember(user_id, message)
    context = await memory.acontext_block(user_id, message)
    return {"context": context}

Inject into system prompt

context = memory.context_block("user123", query="user preferences", max_tokens=500)
# Returns formatted string ready for system prompt injection

GDPR-compliant deletion

memory.forget("user123")               # Delete all memories for user
memory.forget("user123", memory_id)    # Delete one specific memory

How it works

kemi sits between your agent and your storage. It handles:

• Semantic deduplication — "I'm vegetarian" and "I don't eat meat" are the same memory
• Importance-weighted scoring — recent, important memories rank higher
• Temporal decay — memories fade if never recalled
• Conflict detection — flags contradictory memories for review
• Lifecycle management — active → decaying → archived → deleted

Adapters

Type	Default	Available
Embedding	fastembed (local)	OpenAI, custom
Storage	SQLite	JSON, custom

Documentation

• Quickstart — get running in 5 minutes
• Recipes — complete working examples
• Configuration — tuning kemi for your use case
• Adapters — embeddings, storage, and custom implementations

License

MIT — free forever, no exceptions.