npcpy

What is npcpy?

npcpy is a Python library that provides key primitives for research and development with multimodal language models, agentic AI, and knowledge graphs. It supports both local providers (ollama, llama.cpp, onnx, LM Studio) and cloud providers, and enables building multi‑agent teams with a Context‑Agent‑Tool data layer that enforces compliance through software rather than prompts.

How to use npcpy?

Install with pip install npcpy. Then import the library to create NPC personas, run direct LLM calls, build agents with built‑in or custom tools, orchestrate multi‑agent debates via NPCArray, or manage knowledge graphs with sleep/dream lifecycle functions. Example code is provided for each major feature.

Key features of npcpy

• Primitives for multimodal LLMs, agentic AI, and knowledge graphs
• Support for local and cloud model providers
• Multi‑agent debate and orchestration via NPCArray
• Context‑Agent‑Tool data layer for safety compliance
• Pre‑built agent tools: shell, Python, edit_file, web_search
• CodingAgent for automated code execution

Use cases of npcpy

• Build multi‑agent teams to debate and reach consensus on complex problems
• Create persona‑based NPCs for interactive storytelling or simulations
• Fine‑tune diffusion models with agent‑driven data collection and training
• Simplify context engineering for large language model applications
• Generate and execute code with the auto‑executing CodingAgent

FAQ from npcpy

What is the NPC Context-Agent-Tool data layer?

It is a framework within npcpy that structures interactions between personas (NPCs), agents, and tools to ensure compliance with directives through software rather than relying solely on prompts.

How do I install npcpy?

Run pip install npcpy in your environment.

What model providers does npcpy support?

npcpy supports local providers such as ollama, llama.cpp, onnx, and LM Studio, as well as cloud providers (e.g., ollama’s cloud models).

How do I run multi-agent debates with npcpy?

Use the NPCArray class to create a team of NPC objects, then call team.infer() with a prompt, optionally followed by chain() for iterative refinement. The README includes a full example with role‑based personas and debate rounds.