Our take
Inactiveno commits in over a year- Activity
- 19
- Community
- 51
- Issues
- 21
- Pull requests
- 20
+2 stars in the last 30 days
Memary provides a concrete, integrated design for agent memory, but its lack of recent development makes it better suited to evaluation or adaptation than as an actively maintained dependency.
Good fit if
- You want to study or adapt a Python example combining LlamaIndex, graph retrieval, and entity-based memory.
- You can work within its Python version range and configure the model and graph services it uses.
- You are comfortable reviewing an MIT-licensed project that has several contributors and is not concentrated around one person.
Look elsewhere if
- You need a project with recent commits, releases, or visible recent issue and pull-request handling.
- You need tests or CI as part of the repository's stated setup, since neither is reported.
- You need the package to support any agent provider without adapting it. The README says removing the included ReAct agent to support agents from any provider was future work.
All health signals
| Last commit | 2024-10-22 (2 years ago) |
|---|---|
| Commits, last 90 days | 0 |
| Releases, last 12 months | 0 (latest v0.1.5, 2024-10-22) |
| Contributors | 7 (top contributor: 47% of commits) |
| Issues closed, last 90 days | 0 |
| Pull requests merged, last 90 days | 0 |
| Project age | 2 years |
Checked on 2026-10-25 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
2,663Forks
207Change since Nov 17, 2026: 0 stars·0 forks
Overview
Memary is an open-source memory layer intended to give autonomous agents a record of what users discuss and know. It is built around LlamaIndex knowledge-graph retrieval and a ReAct chat agent, with the README describing a graph-backed knowledge store alongside memory modules that track entity mentions over time. The project frames this memory as a way to help agents reason with relevant context and adapt responses to users' interests and expertise.
It is aimed at developers building Python agents who want an integrated example of graph retrieval, interaction memory, and configurable tools. The included setup uses local Ollama models by default, with OpenAI model options also described; graph database configuration can use FalkorDB or Neo4j.
Key Features
- Extracts entities from interactions and writes agent responses back into a knowledge graph.
- Uses recursive retrieval to find relevant graph context, with multi-hop handling for queries involving multiple entities.
- Maintains a memory stream of entities and timestamps, plus an entity store tracking mention frequency and recency.
- Summarizes older chat history to manage context as conversations grow.
- Provides a ReAct-based
ChatAgentwith default search, vision, location, and stock tool options. - Supports adding and removing custom tools, and separate agent contexts through FalkorDB multi-graphs.
- Allows configuring local Ollama models or the listed OpenAI model options.
Use Cases
- A Python agent developer wants a starting point for combining conversation history with knowledge-graph retrieval.
- A personal-assistant builder wants to track recurring topics and use them to tailor later responses.
- A team prototyping multiple user-specific agents wants separate memory and graph contexts through FalkorDB.
- A developer wants an example of adding custom tools to a ReAct agent while retaining graph-backed search.
What you need
Detected in the repository
- Python >=3.8, <=3.11.9 (from pyproject.toml)
License in plain words
MITpermissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the copyright and license notice
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
The README requires Python 3.8 through 3.11.9. Install the package with:
pip install memary
For model, database, and credential setup, see the README and documentation.
Alternatives
- open-index: Open Index uses a structured context graph with SQLite or OpenSearch and MCP, rather than Memary’s LlamaIndex-based knowledge graph and ReAct agent.
- OpenViking: OpenViking organizes agent memories, knowledge, and skills in a browsable virtual filesystem with scoped retrieval.
- ReMe: ReMe uses a local-first, self-evolving Markdown knowledge base for capturing and retrieving durable context.
- taOS: taOS provides memory as part of a self-hosted agent platform with chat and files, rather than as a standalone memory layer.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| Memary | Jupyter Notebook | MIT | 2.7k | Inactive |
| open-index | Python | MIT | 112 | New |
| OpenViking | Python | AGPL-3.0 | 39.2k | Active |
| ReMe | Python | Apache-2.0 | 3.7k | Active |
| taOS | Python | AGPL-3.0 | 603 | Active |
Considerations
- The repository is inactive: it has had no commits for over a year, no releases in the past year, and no recent issue closures or merged pull requests. Its latest release is from the same period as its last commit.
- The reported runtime range ends at Python 3.11.9. The README also expects model configuration and API credentials for relevant external services; graph database use involves FalkorDB or Neo4j configuration.
- The repository reports no tests or CI, so adopters should validate installation and behavior in their own environment.
- The included ReAct agent is described as a demonstration implementation. The README says broader support for agents from any provider is planned, not already delivered.
- The project uses the MIT license; review its terms when incorporating or distributing it.
Repository: kingjulio8238/Memary.
Found this useful?
Share it with someone who would like Memary.