OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Memory & RAG

Plugmem

PlugMem is an open-source Python long-term memory library for LLM agents. It structures experience into compact semantic, procedural and episodic knowledge units, links them in a memory graph, and supports retrieval for agent reasoning.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Plugmem — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use PlugMem when an agent needs compact, reusable context from prior interactions and you can integrate a Python memory component into its pipeline. The repository’s benchmark examples are relevant when preparing or evaluating such memory-backed agents.

Who it is for

  • Developers adding persistent memory to an LLM-agent pipeline.
  • Teams building dialogue, knowledge-intensive question-answering or web-automation agents.
  • Researchers working with agent-memory benchmarks such as WebArena, LongMemEval or HotpotQA.

Documented uses

  • Save experience as reusable knowledge rather than depending on raw interaction histories.
  • Retrieve relevant facts, procedures or past interaction sequences for a later agent decision.
  • Load a persisted memory graph for evaluation, use it read-only, refresh its embeddings or disable graph operations in documented WebArena examples.
  • Adapt memory behavior by defining value functions and reasoning prompts.
Before you choose

Requirements & limits

The public GitHub repository specifies an Apache-2.0 license.

  • The library uses a graph of hierarchical knowledge units and distinguishes semantic, procedural and episodic memory.
  • The documented quick start depends on configured OpenAI and Azure parameters, local Qwen and embedding services, and a structured data directory.
  • Benchmark reproduction requires additional benchmark environments and may involve benchmark-specific changes.
  • The README describes an installable Claude Code plugin and an OpenClaw plugin integration; these are integrations around the library rather than a replacement for its role as an agent-memory component.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare PlugMem’s documented quick-start environment

Set the parameters, local inference services and memory directories required before running a PlugMem benchmark example.

Before you start

  • A local PlugMem checkout and a data path for its memory files.
  • An OpenAI API key and Azure endpoint.
  • Hosts for the Qwen and embedding services, with the documented Qwen and NV-Embed-v2 deployment scripts.
  • The benchmark environment setup described in the repository, including openai==2.6.1.

Steps

  1. Set OPENAI_API_KEY, AZURE_ENDPOINT, DIR_PATH, QWEN_BASE_URL and EMBEDDING_BASE_URL to the values for your environment.

  2. From the repository, run the Qwen and NV-Embed-v2 deployment scripts in host_local_inference.

  3. Create episodic_memory, semantic_memory, procedural_memory, tag and subgoal directories under DIR_PATH.

Expected outcome

The documented environment variables, inference services and memory directory structure are in place for proceeding to a benchmark example.

Guide limitations

  • These steps prepare the environment; they do not run an evaluation.
  • Benchmark installation and any benchmark-specific modifications are separate prerequisites.
  • The README’s sample service URLs use placeholders that must be replaced with your own host values.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
TIMAN-group/PlugMem · Official homepage title · retained in its source language
How we review sourcesSuggest a correction