DOCSEARCHbyAlgolia
Submit Docs

Naptha AI

website·Indexed 33 days ago

https://docs.naptha.ai

A platform for building and managing AI agents, tools, knowledge bases, and memories. It provides a comprehensive SDK for node management, decentralized storage via IPFS, gRPC API services, and an inference interface for various models.

Tokens
59K
Snippets
383
Records
446
Agent score
61%

What's inside Naptha

  1. Overview of Naptha Memory Modules

    Memory modules in Naptha allow agent modules to store, retrieve, and manage experiences to maintain context and learn from past interactions. They are deployed on Naptha Nodes, with state stored in a local PostgreSQL database and the file system. Supported memory types include:

    • Cognitive Memory: For reflections and learned insights.
    • Episodic Memory: For recording sequential experiences.
    • Semantic Memory: For storing knowledge and facts.

  2. Overview of Naptha Module types

    Naptha Modules are framework-agnostic building blocks for multi-agent applications that can run on separate devices and interact over a network. There are seven primary types of modules:

    • Agent Modules: Chat, Task-solving, or ReAct agents.
    • Tool Modules: Web Search, Python Code Execution, etc.
    • Knowledge Base Modules: Wikipedia, Google Search, etc.
    • Memory Modules: Long-term and Short-term memory.
    • Persona Modules: Social or synthetic market personas.
    • Agent Orchestrator Modules: Organizations of agents or social simulations.
    • Environment Modules: Group chats, information boards, or auctions.

  3. Overview of Naptha Tool Modules

    Tool modules provide reusable functions that extend the capabilities of agents. They allow agents to perform specialized tasks such as web searching, database querying, file searching, or executing custom functions. Because Tool.run calls worker nodes via API, agents and tools can interoperate even if they are built using different frameworks or languages.

  4. Overview of Naptha Environment Modules

    Environment modules in Naptha provide the infrastructure necessary for reinforcement learning. Examples include group chats, information boards, job boards, social networks, and auctions. These modules are deployed on Naptha Nodes, and their state is persisted using a local PostgreSQL database and the file system.

  5. Overview of Knowledge Base Modules

    Knowledge Base (KB) modules provide agents and other modules with access to structured information repositories. They allow agents to query, retrieve, and reason about domain-specific knowledge, such as specialized domain knowledge, message history (conversation logs), or technical documentation. KB modules are deployed on Naptha Nodes, with state stored in a local Postgres database and the file system.

  6. Overview of Naptha Storage features

    Naptha Storage provides a secure way for Naptha Modules to store and retrieve data locally and privately on Naptha Nodes. It features a consistent interface across all storage types supporting basic operations: create, read, update, delete, list, and search. The system is extensible via the StorageProvider interface and uses an options pattern to support storage-specific features without breaking the common API.

  7. Overview of the Naptha Multi-Agent Framework

    Naptha is a framework and infrastructure for developing and running multi-agent systems at scale. Unlike monolithic AI systems, Naptha enables a network of heterogeneous agents—using different models, architectures, and data—to interact over a network across separate devices. This architecture supports distributed workflows that can run on one or more Naptha nodes, allowing for increased performance through model diversity (heterogeneity) and scalability via ensembles of smaller LLMs. Key advantages include enhanced privacy (secure use of proprietary data) and improved system safety (robustness against individual agent failure).

  8. Overview of Naptha Node services

    The Naptha Node packages several services to support agentic workflows. These include:

    • Local Inference: Supports VLLM or Ollama for local model inference, with specialized support for tool calling in select open source models.
    • LiteLLM Proxy Server: Provides a unified OpenAI-compatible API interface for multiple LLM providers.
    • Local Server: An HTTP, Web Sockets, or gRPC server that allows other agents in the network to access published modules via API.
    • Local Storage: A Postgres database and file system used to store the state of Environment modules (e.g., group chats, job boards) and inference metadata (token usage, costs).
    • Module Manager: Handles the installation of agents, tools, orchestrators, environments, and personas from GitHub, HuggingFace, and IPFS.
    • Message Broker and Workers: Uses RabbitMQ for asynchronous processing between modules, executed via Poetry or Docker.
    • Local Hub (Optional): A local registry for modules and nodes using SurrealDB, enabled by setting LOCAL_HUB=true in the .env file.

  9. Overview of Naptha Decentralized AI Orchestration

    Naptha is a framework and infrastructure designed to enable decentralized AI orchestration. Unlike traditional cloud workflow orchestrators (such as Airflow, Kubeflow Pipelines, or Temporal) that are designed for single-node execution, Naptha allows decentralized workflows to run across one or more nodes. This architecture supports 'true' multi-agent systems where agents can utilize different LLMs and access various local data sources, addressing scalability and data privacy issues associated with centralized LLM application hosting.

  10. Generate Synthetic Data using LLM Frameworks

    Naptha can be used to build data frameworks that leverage LLMs for the following tasks:

    • Curating existing datasets
    • Processing raw data
    • Generating new training data
    • Synthesizing edge cases

  11. Understand Naptha Modules for multi-agent systems

    Naptha Modules are framework-agnostic building blocks used to create multi-agent systems. They allow developers to implement components using different agent frameworks while maintaining interoperability. There are seven primary module types: Agents, Tools, Knowledge Bases, Memories, Personas, Orchestrators, and Environments.

  12. Understand Naptha Orchestrator Modules

    Orchestrator modules manage the coordination of agents, tools, environments, and personas based on specific interaction patterns and workflows. Common use cases include social simulations (debates), task-solving swarms (like BabyAGI or MetaGPT), and market prediction agents. The core orchestration logic is typically implemented in the run.py file of the module.