npcpy

Npcpy is a versatile, Python‑centric framework that turns large language models into fully fledged agents capable of interacting with the world through tools, templates, and coordinated teams. By exposing a simple yet expressive API for defining agents, tool calls, and Jinja‑based workflows, it solves the common pain point of integrating LLMs into production pipelines: how to make an assistant do real work while keeping the model’s reasoning transparent and controllable.

At its core, Npcpy lets developers declare an Agent by specifying a name, a primary directive (the agent’s purpose), the underlying model and provider, and an optional set of Python functions that serve as tools. The framework automatically handles the orchestration of tool calls, including formatting arguments, executing the function, and feeding results back into the model’s conversation. This eliminates boilerplate code for parsing LLM output, managing state, and handling errors, enabling rapid prototyping of assistants that can read files, query databases, or invoke external APIs without the model “hallucinating” unsupported operations.

Beyond single agents, Npcpy introduces Team and Jinx constructs. A team aggregates multiple agents with distinct roles, allowing them to collaborate on complex tasks—much like a human squad where each member brings specialized knowledge. Jinx templates are prompt‑driven pipelines that can be executed entirely through prompts, making them model-agnostic. This is particularly valuable when working with models that lack native tool‑calling capabilities; the Jinja engine can still orchestrate external logic by embedding Python code blocks and rendering results into subsequent prompts.

The practical impact is significant for developers building AI‑powered applications. Use cases include automated code review, data analysis pipelines that read and summarize large datasets, or content generation workflows where an assistant composes markdown documents after querying a file system. Because Npcpy manages the full conversation history and tool call lifecycle, developers can focus on defining business logic rather than handling low‑level LLM interactions.

In summary, Npcpy offers a high‑level abstraction that bridges the gap between raw language models and real‑world applications. Its tool‑centric design, team coordination features, and Jinja execution engine give developers a powerful, flexible toolkit for building reliable, explainable AI assistants that can perform concrete tasks while maintaining clear traceability of each step.