It introduces a novel evidence-organization step (ledger) between retrieval and generation, and provides a one-command automatic memory wrapper for OpenAI clients, with visual demos and benchmark case replay.
ShadowWeave Team
The Holographic Memory System (HMS) is a structured long-term memory layer for AI applications. It retains conversations and documents, extracts durable facts, links related entities and events, and recalls relevant context for later model calls.
HMS is designed for applications that need memory across sessions without placing an entire conversation history into every prompt.
HMS can wrap an existing OpenAI client so each model call automatically:
user input -> recall relevant memories -> inject context -> call the LLM
-> retain the completed user/assistant exchange
Configure the model Base URL, API key, and model in .env, then run:
bash scripts/run_memory_demo.sh
The script starts PostgreSQL and HMS locally, waits for the memory API, installs
the local SDK adapter in an isolated environment, and runs a two-turn demo. The
first turn stores a user preference and project; the second turn recalls both
without manually calling retain() or recall().
The application-side integration is one wrapper call:
from openai import OpenAI
from hms_litellm import wrap_openai
client = wrap_openai(
OpenAI(),
hms_api_url="http://127.0.0.1:18080",
api_key="YOUR_HMS_API_KEY",
bank_id="user-alice",
)
response = client.responses.create(
model="gpt-4o-mini",
input="What do you remember about my current project?",
)
wrap_openai() supports both client.responses.create(...) and
client.chat.completions.create(...), including streaming. Use a stable,
per-user bank_id; optionally set session_id to accumulate one conversation
as a tracked HMS document.
The dataplane includes an opt-in openai_multimodal file parser. Images are
validated and normalized locally; videos are decoded locally into a bounded,
deterministic frame set. The visual description is rendered as grounded
canonical Markdown and then enters the existing document, chunk, embedding,
link, and recall pipeline. Raw video is never sent to the description provider.
The feature is disabled by default. Its current runtime support matrix is PostgreSQL; enabling the media path
Similar projects matched by category, topics, and programming language.
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.