title	emoji	colorFrom	colorTo	sdk	python_version	sdk_version	app_file	pinned	short_description	tags
Aware NPC chat	💬💡	indigo	gray	gradio	3.13	5.31.0	src/app.py	false	A chat interface with an NPC aware of the world it lives in	agent-demo-track

Chat with aware NPCs

I love roleplaying and AI Agent may bring a whole new experience to chatting in games.

I want to build an AI Agent system that takes allow NPCs (Non Player Characters) to talk with you according to their own knowledge of the world they live in.

Program

• Prerequisites:

  • Build a knowledge base (characters, places, events, objects)
  • Build a simple and effective chat interface
  • Basic agents to retrieve useful information and converse

• Iterate to improve quality:

  • Improve the agent system, allow for personalization of the responses
    • Provide a PersonalizedAgent over the CodeAgent and MultiStepAgent to allow more flexibility in initial state variables and system prompt templating
    • Remarkable changes on the prompts to fit the thinking process of a character and not a robotic agent
  • Integrate an emotional dimension into the responses
    • Through the planning template
  • Integrate a notion of willingness to answer
    • Through the planning template
  • Grade the level of accessible knowledge

• Bonus, not everything can be done:

  • Generate alternative images translating the character's emotions with each response
  • An interface "behind the scenes of the chat," to visualize the work of the agents in the background
  • Generalize the process of building the knowledge base

Install the project

Requires Python 3.13 or higher.

Install steps, run refers to command lines in your terminal :

1. Clone the GitHub project
2. You might need to download Git LFS in order to access the static large files. Then run git lfs install and finally git lfs pull.
3. Create a virtual environment (recommended)
4. Install uv dependency manager e.g. run pip install pipx then pipx install uv in a terminal
5. Run uv sync then uv pip install -e .
6. Set up your secrets (like HF_read_token), see example list in the .example.env file.
7. Run the app with uv run src/app.py

Build a knowledge base from a fandom website

Download the XML file of the Fandom website :

from knowledge_base.utils.downloader import download_file, fetch_page_content, get_xml_dump_url
from knowledge_base.utils.url import get_fandom_statistics_page_url

XML_PATH = r"....\asimov_fandom_dump.xml"

fandom_url = "https://asimov.fandom.com/wiki/"
statistics_url = get_fandom_statistics_page_url(fandom_url)
statistics_page_content = fetch_page_content(statistics_url)
xml_dump_url = get_xml_dump_url(statistics_page_content)

xml_file = download_file(xml_dump_url, output_path=XML_PATH)

From xml dump to knowledge base :

from knowledge_base.models.knowledge_base import KnowledgeBase
from knowledge_base.parser.fandom.bridge_site_to_kb import populate_entities, populate_relationships
from knowledge_base.parser.fandom.parse_dump import fandom_xml_parse

XML_PATH = r"....\asimov_fandom_dump.xml"  # <= Same as previous code blob
KB_PATH = r"....\kb_asimov.json"

fandom_site_content = fandom_xml_parse(XML_PATH)

kb = KnowledgeBase()
populate_entities(fandom_site_content, kb)
populate_relationships(fandom_site_content, kb)
kb.save_kb(KB_PATH, compress=True)  # <= Compressing automatically add the .gz extension