MagiCore is a standalone .NET memory library that runs locally with the providers you configure.
- Follow Getting started to install the package and learn the memory lifecycle.
- Run the sample projects, beginning with the zero-setup console application.
- Use the API reference when you need filters, scopes, paging, expiration, spatial and robotics memory, or extension interfaces.
| Goal | Guide |
|---|---|
| Build and test without external services | Getting started |
Use OpenAI, Azure OpenAI, Ollama, or another Microsoft.Extensions.AI provider |
Providers and persistence |
| Persist vectors in PostgreSQL, SQLite, Qdrant, or another vector store | Providers and persistence |
| Build an embodied 3D memory workflow | Godot spatial memory sample |
| Add reranking or custom providers | Providers and persistence |
| Expose memory through local MCP tools | Getting started |
- API reference describes public contracts and configuration models.
- Architecture explains dependency direction and extension boundaries.
- Evaluation describes the benchmark harness and the latest measured results.
- Contribution guide covers local development and pull requests.
new MemoryService() uses in-memory storage, deterministic lexical hashing embeddings, and basic message extraction. This path is useful for development and tests, but it is not a semantic-quality baseline or durable production storage. Choose model-backed embeddings and a persistent store for production workloads, and keep the configured embedding dimensions identical across the provider and vector store.