Arkhon Memory SDK

Clean, LOCAL (!!!), long-term memory for autonomous LLM agents and agentic systems.
A foundational component built to support persistence, learning, and context recall — without database bloat or framework lock-in.

📦 Install

pip install arkhon-memory

Supports Python 3.8+ and pydantic 2.x.

🚀 What is Arkhon Memory?

Most agent “memory” is brittle: chat logs, hacky context, or heavyweight vector DBs.

Who is this for: Indie hackers, researchers, agent devs, and tinkerers, ANYONE tired of vector DBs or lock-in memory “solutions”.

Builders who want cognitive-like, composable memory — not chat history hacks - AND all of it is happening/stored locally not in anyone elses cloud.

Arkhon is a lightweight, composable memory layer:

• JSON-native, structured, and easy to reason about
• Tracks relevance over time (with time decay + reuse boosting)
• Designed for event-driven, session-aware agent workflows
• No dependencies except Python & pydantic

Arkhon is the “active memory” foundation of a much larger system we call Cathedral.
This SDK is for builders who want real persistence — and want to keep their stack simple.

✨ Features

• Plug-and-play: add persistent memory to any LLM agent in minutes
• Time decay & reuse: fresher and more-used facts stay relevant
• Tagging & snapshots: organize, search, and summarize memory easily
• Session lifecycle hooks: manage memory at start/end of sessions
• No bloat: no vector DBs, no LangChain, no black-box magic

📄 License

MIT — free to use, modify, or integrate.

🤝 Contributing & Feedback

We welcome feedback, issues, and pull requests!
If you find a bug or have ideas for improvement, please open an issue or submit a PR on GitHub.

For questions, feature requests, or direct contact:
Email: kissg@me.com

Star the repo if you find Arkhon useful or want to follow updates!

🛠️ Quick Start (included in examples folder)

from arkhon_memory.memory_hub import MemoryHub
from arkhon_memory.schemas import MemoryItem
from arkhon_memory.lifecycle import on_session_start, on_session_exit
from datetime import datetime

hub = on_session_start("demo_memory.json")
item = MemoryItem(
    content="Tokyo has the world's largest metro population.",
    tags=["geography", "asia"],
    timestamp=datetime.utcnow(),
)
hub.append(item)

results = hub.query("tokyo")
for r in results:
    print("Found:", r.content, r.tags)

on_session_exit(
    hub,
    tags=["demo", "test"],
    summary="Demo session saving basic fact about Tokyo.",
    title="Tokyo Fact Demo"
)