Documentation Chatbot

NeuralMemory includes a self-answering documentation chatbot powered by spreading activation — no LLM required.

Try it live: HuggingFace Space

How it works

1. Project documentation is encoded into a neural memory brain (neurons + synapses + fibers)
2. Your query triggers spreading activation across the knowledge graph
3. The most relevant documentation chunks are retrieved and displayed
4. A confidence score reflects how well the context matches your query

The chatbot uses ReflexPipeline — the same retrieval engine behind nmem_recall.

Running locally

pip install neural-memory gradio
python chatbot/app.py

Options:

Flag	Description
--port 7861	Custom port (default: 7860)
--share	Create a public Gradio URL

Re-training the brain

If you've updated the documentation:

python chatbot/train_docs_brain.py

This trains from docs/, README.md, CHANGELOG.md, and FAQ.md. The brain is saved to chatbot/brain/docs.db.

Training options:

Flag	Description
--brain NAME	Custom brain name (default: neuralmemory-docs)
--export DIR	Copy the trained DB to another directory
--no-verify	Skip verification queries

Deploying to HuggingFace Spaces

Prerequisites

pip install huggingface_hub
huggingface-cli login

One-command deploy

bash chatbot/deploy_hf.sh
# or with a custom Space name:
bash chatbot/deploy_hf.sh my-org/my-space

Manual deploy

1. Create a new Gradio Space with SDK = Gradio
2. Clone the Space repo
3. Copy chatbot/app.py, chatbot/requirements.txt, chatbot/README.md, and chatbot/brain/ into the Space
4. Push to HuggingFace

The brain DB is ~51 MB — well within HuggingFace's file size limits.

Search depth levels

Level	Pipeline Depth	Speed	Best for
Quick	INSTANT	~5ms	Simple keyword lookups
Normal	CONTEXT	~20ms	Most questions
Deep	DEEP	~50ms	Complex multi-topic queries