**Modular AI Framework for Context-Aware, Goal-Driven Agents**
Neucore is a TypeScript framework designed for building intelligent, context-aware AI applications with memory, reasoning, and action capabilities. It follows a modular architecture, supporting features like vector-based memory retrieval, structured reasoning (e.g., Chain of Thought), context builders, and plugin-based AI provider adapters. It introduces the **Model Context Protocol (MCP)** to structure interactions between user intents and model responses, facilitating composability and multi-agent systems.
**Key Features:**
- Memory engine with vector embeddings for semantic recall
- Reasoning system supporting Chain of Thought and planned multi-step logic
- Context builder to dynamically assemble relevant input for prompts
- Action system for executing validated operations based on AI intent
- Character system for defining consistent agent behavior and traits
- Provider adapters for Anthropic (with OpenAI support planned)
- Built-in RAG support for combining retrieval with generation
- Extensible and type-safe design using modern TypeScript patterns
**Example Use Cases:**
- Build intelligent agents with goals and planning capabilities
- Create reasoning pipelines that combine memory, context, and task execution
- Route AI intents to specific model handlers using the MCP abstraction
**Tech Stack:** TypeScript, Node.js, Bun (optional), Embedding search, ZK-style modularity
**License:** MIT
[View on GitHub](https://github.com/ngmachado/neucore)
## Related Notes
- [[202504121408 - Chain of Thought Reasoning]]
- [[202504121408 - Model Context Protocol]]
- [[202504121408 - AI Memory Systems]]
- [[202504121408 - Context-Aware AI Applications]]
- [[202504121408 - Reasoning Methods in AI]]