HOMEBASE

Multi-Agent Home Management System — LangGraph + Groq + Gemini + Anthropic

v1.19.0 | 📖 Documentation

⚠ Proof of Concept — Not Production Ready HOMEBASE is a demonstration system built to illustrate multi-agent agentic AI architecture patterns. It has not undergone formal code review, security assessment, penetration testing, secrets management audit, or production hardening. It should not be deployed in a production environment, used to process real sensitive data, or presented as a production-grade system without a full security review, compliance evaluation, and architectural assessment appropriate to the target environment and regulatory context.

A multi-agent system built with LangGraph, Groq (Llama 3.3 70B), Gemini (2.5 Flash-Lite), and Anthropic (Claude Sonnet) demonstrating orchestrator/subagent delegation, parallel agent execution, multi-provider LLM routing, human-in-the-loop (HITL) checkpoints, and state persistence. The domain is home management; the architecture is enterprise-transferable.

---

Quickstart

Requirements: Python 3.11+, uv

git clone https://github.com/ianmeinert/homebase.git
cd homebase
uv sync --dev
cp .env.example .env
# Add GROQ_API_KEY to .env — get one at https://console.groq.com
uv run streamlit run app.py

For CLI usage, demo data seeding, LangSmith tracing setup, and Google API key configuration see the Setup guide.

---

Architecture

orchestrator  (trigger-based category filter + optional HU/HI-only mode)
    +-- hvac_agent        -+
    +-- plumbing_agent     |
    +-- electrical_agent   +- (parallel fan-out, one Groq call per agent)
    +-- appliance_agent    |
    +-- general_agent     -+
            |
      hitl_briefing        <- graph pauses here (interrupt_before synthesizer)
            |
       [human input]       <- approve / defer HU/HI + LU/HI items / add notes
            |
       synthesizer         <- Claude Sonnet or Groq generates narrative (runtime provider selection)
            |
           END

Full architecture docs: Architecture Overview | Multi-Provider Strategy | Data Model

---

Tests

uv run pytest

No API key required — all LLM calls are mocked. 619 passing tests across 18 files.

---

Documentation

Section	Description
Setup	Install, env vars, demo data, LangSmith
Running	UI features, CLI, prompt library, test table
Architecture	Graph topology, project structure
Agents	Agent reference, LLM vs rule-based breakdown
Enterprise Bridge	Stakeholder concept map
Changelog	Version history
Backlog	Feature backlog