Open Memory Protocol launched on GitHub as a shared memory layer for Claude, ChatGPT and Curso. The project surfaced in an HN thread that reached 26 points and 9 comments.
What It Is / How It Works
The protocol stores conversation state in one backend that multiple models can read and write. Each supported model connects through a thin adapter instead of maintaining separate memory files.
Adapters translate model-specific message formats into a common schema. The store itself uses standard key-value operations with optional vector search for retrieval.
How to Try It
Clone the repository and install the Python package. Point each model's client to the same endpoint and memory namespace.
git clone https://github.com/SMJAI/open-memory-protocol
pip install open-memory-protocol
Run the server locally, then configure Claude and ChatGPT clients with the provided adapter URLs.
Pros and Cons
- Single source of truth reduces duplication across sessions
- Works with three major chat interfaces out of the box
- Requires running an additional service
- No public benchmarks on retrieval latency or token savings yet
- Early HN comments note limited documentation for production use
Alternatives and Comparisons
LangChain memory modules and Mem0 both offer persistent storage, but each ties closely to one framework. Open Memory Protocol focuses on cross-model compatibility without requiring a full agent framework.
| Feature | Open Memory Protocol | LangChain Memory | Mem0 |
|---|---|---|---|
| Cross-model support | Yes | Partial | No |
| Local server option | Yes | Yes | Yes |
| HN discussion points | 26 | N/A | N/A |
| License | Not specified | MIT | Apache 2.0 |
Who Should Use This
Developers who switch between Claude, ChatGPT and Curso in the same workflow gain the most. Teams already invested in a single framework such as LangChain see smaller benefits.
Skip it if you need only one model or already run a mature memory layer with measured performance numbers.
Bottom Line / Verdict
Open Memory Protocol offers a lightweight way to keep one memory store across three chat interfaces, backed by a 26-point HN thread and a public GitHub repository. Early adopters can test it in under ten minutes using the provided adapters.
Top comments (0)