ryoma-nicer 0.8.0

a clone from the root project for Ryoma AI Platform – made compatible with Pydantic v2

Ryoma Nicer

A Pydantic v2–compatible fork of Ryoma AI Platform

---

Overview

Ryoma Nicer is a community-maintained fork of the Ryoma AI Platform, originally created by the Ryoma team. The upstream project provides an AI-powered data agent framework for seamless data analysis, engineering, and visualization.

This fork is:

• Pydantic v2–compatible (ryoma-nicer runs on Pydantic 2.x)
• LangGraph-ready, updated for modern LangGraph (StateGraph + Pregel) APIs
• Actively maintained as part of the “NICER” suite of agents (e.g. baby-NICER)

It is used in production within the NICER stack to power agents such as WorkflowAgent and SQL agents that run on a LangGraph / LangChain / LangMem tooling ecosystem.

For the original project, see project-ryoma/ryoma. All architectural credit for the base design goes to the upstream maintainers.

Why This Fork?

Pydantic v2 introduced breaking changes in how models are defined and validated. Many projects, including the original Ryoma AI Platform, were tightly coupled to Pydantic v1. To allow developers to adopt the latest Pydantic improvements without sacrificing Ryoma functionality, this fork:

1. Migrates all core model definitions to the Pydantic v2 BaseModel API.
2. Updates validation and config patterns to leverage v2’s faster runtime and stricter type checking.
3. Aligns with a modern agent stack, including:
   • LangGraph (tested with langgraph==0.6.x)
   • LangChain / LangChain Core (tested with langchain==0.3.x, langchain-core==0.3.x)
   • Usage inside the NICER suite of agents (e.g. baby-NICER / langgraph-slack).

Installation

Install the Pydantic v2–compatible release from PyPI:

pip install ryoma-nicer

Optionally, include supported data source extras:

pip install "ryoma-nicer[snowflake,pyspark,postgres,sqlite,mysql,bigquery]"

Usage

The API surface and usage mirror the original Ryoma. For example, to run a simple SQL agent:

from ryoma_ai.agent.sql import SqlAgent
from ryoma_ai.datasource.postgres import PostgresDataSource

# Initialize data source
datasource = PostgresDataSource("postgresql://user:pass@host:5432/db")

# Create and run SQL agent
agent = SqlAgent("gpt-3.5-turbo").add_datasource(datasource)
agent.stream("SELECT count(*) FROM orders", display=True)

For conceptual details and additional examples, the upstream Ryoma documentation still applies in most cases. Where this fork differs is primarily in:

• Pydantic v2 model definitions
• Updated integrations used by the NICER agent stack

Usage in the NICER agent suite

Within the NICER ecosystem (e.g. baby-NICER / langgraph-slack), this fork powers agents that run on LangGraph’s compiled graphs (StateGraph.compile(...) returning a Pregel runtime). For example, WorkflowAgent:

• builds a StateGraph over MessageState
• compiles it to a Pregel runtime with MemorySaver checkpointers
• uses LangGraph’s invoke/stream APIs to manage tool-augmented workflows

This ensures ryoma-nicer works cleanly with modern LangGraph-based systems while preserving the spirit of the original Ryoma design.

Contribution & Upstream

• Forked from: project-ryoma/ryoma
• Upstream credit: Architectural ideas, core abstractions, and much of the original code come from the Ryoma team.
• Issue tracker & pull requests: Please use this repository to report Pydantic v2 / modern-stack compatibility issues or to propose improvements needed by the NICER suite of agents.

License

This project is released under the Apache License 2.0.