Apppublic

cygon24/local-ai-lab-v6

sourceHugging Faceupdated 6mo agoView on Hugging Face

0likes

App README

Local AI Lab — V6 Multi-Agent

Part of the Local AI Lab series — building AI systems from scratch, one version at a time.

Version	Focus
V1	Basic Ollama chat
V2	Sessions + context window
V3	RAG + ChromaDB
V4	ReAct agent — pure Python
V4 Extended	Web tools + observability + MCP + memory
V5	LangChain + LangGraph + Phoenix tracing
V6	Multi-agent — Orchestrator + Researcher + Executor

Architecture

User goal
    ↓
🟣 Orchestrator  — analyzes goal, routes, synthesizes final answer
    ↓
🔵 Researcher    — web_search, fetch_webpage, knowledge base
🟠 Executor      — calculate, run_python, read/write files
    ↓
🟣 Orchestrator  — final answer

Setup on HF Spaces

Required secret:

GROQ_API_KEY = your_key_here

Add it in: Space Settings → Variables and secrets → New secret

Get a free key at: https://console.groq.com (no billing required)

Running locally

bash

# Set your Groq key
export GROQ_API_KEY=your_key_here

# Install
pip install -r requirements.txt

# Run
streamlit run app.py

Or use Ollama without a Groq key:

bash

ollama serve
streamlit run app.py

Stack

LangGraph          multi-agent graph
LangChain          tool decorator, ChatGroq
Groq API           free hosted LLM (llama-3.1-8b-instant)
sentence-transformers  embeddings (no GPU needed)
ChromaDB           vector store — persistent on /data
SQLite             long-term memory — persistent on /data
Streamlit          UI

GitHub: github.com/cygon23

Built in Arusha, Tanzania