Nerif Architecture
This part, I will introduce core design part by part, here is a general map
Let me help improve the documentation to make it more polished and comprehensive:
Nerif Agent
Like other multi-agent frameworks, Nerif Agent provides flexibility in utilizing various AI models, including multi-modal capabilities. Our framework supports models that can interact with external APIs and tools for enhanced functionality.
Currently, we support fundamental AI capabilities including:
- LLM chat models
- Vision models
- Embedding models
In upcoming releases, we plan to expand support for custom models and external API integrations.
Nerif Core
The key distinction between agent and core lies in their type system implementation. While LLM/VLM models typically generate natural language outputs that require complex post-processing, Nerif Core ensures the outputs are properly typed and immediately usable in your applications.
Our core functionality consists of six essential modules:
- Nerif: Evaluates statements and returns boolean values (
True/False) - Nerification: Validates statements with boolean responses (
True/False) - Nerif Match: Takes a statement and a list as input, returning the index of the best-matching item
- Nerif Format: Handles type conversion between different formats
- Nerif Json: Structures outputs in JSON format according to specified requirements
- Nerif Log: Provides comprehensive logging capabilities
Nerif Flow
This feature will be available after the v1.0 release.