GENDOC.AI

Team members

• Tyson Hoang
• Divine Akinjiyan
• Siddhartha Reddy Pullannagari

Activate virtual environment

conda create -n calhack python=3.12
conda activate calhack

Gendoc.ai — local RAG agent + mailbox agents (no Docker, no API keys)

This repo contains a minimal multi-agent setup using Fetch.ai uAgents with a working RAG (retrieval-augmented) medical helper. It talks via the Agentverse mailbox, retrieves credible snippets from a local Chroma vector DB, and returns conditions/specialties plus citations. You can run everything locally without Docker or API keys.

What’s here

Agents (mailbox mode, unique ports): triage, report, scheduling, pharmacy, audit, orchestrator, and rag (this one does retrieval).

RAG stack: Chroma (HTTP server), SentenceTransformers embeddings, simple rules (MeTTa-ready later).

Senders: send.py, send_rag.py to ping any agent by agent1… address.

Requirements

macOS/Linux, Python 3.13 (for now). (If you want MeTTa/Hyperon reasoning today, use Python 3.11; see “MeTTa mode” below.)

Install deps:

cd agents_lab python -m venv .venv source .venv/bin/activate python -m pip install --upgrade pip python -m pip install "uagents==0.22.10" "pydantic>=2.8,<3" "uvicorn>=0.30.1,<1.0" python -m pip install "chromadb[server]==0.5.5" "sentence-transformers==3.0.1"
"beautifulsoup4==4.12.3" "numpy==1.26.4" "requests==2.32.3"

Run Chroma (no Docker)

Terminal A:

source .venv/bin/activate python -m chromadb.cli run --path .chroma --host 127.0.0.1 --port 8003

Terminal B (env to point at the server):

source .venv/bin/activate export CHROMA_MODE=http export CHROMA_HOST=127.0.0.1 export CHROMA_PORT=8003

Ingest a tiny seed index

ensure packages treat agents/ as a package

touch agents/init.py agents/rag/init.py

ingest public WHO/CDC pages (demo only)

python -m agents.rag.ingest

Run the RAG agent (mailbox) and pair once python -m agents.rag.rag_agent

copy the Inspector link from logs, open it, click: Connect → Mailbox → Finish

copy the printed agent1… address

Send a test query python -m agents.rag.send_rag
--to <RAG_AGENT_ADDRESS>
--text "sore throat 3 days, productive cough, mild fever"
--session sess42 --k 6

You should see:

conditions (demo rules → e.g., flu)

specialties (e.g., primary_care)

citations/facts from Chroma (WHO/CDC snippets)

Run the other template agents (optional)

Each in its own terminal, pair once via Inspector (mailbox):

python agents/triage_agent.py # port 8000 python agents/report_agent.py # port 8002 python agents/scheduling_agent.py # port 8003 python agents/pharmacy_agent.py # port 8004 python agents/audit_agent.py # port 8005 python agents/orchestrator_agent.py # port 8006

Then use the universal sender:

python agents/send.py --to <AGENT_ADDR> --text "..." --target triage|report|schedule|pharmacy|audit

MeTTa (Hyperon) mode (optional, for full symbolic reasoning)

Use Python 3.11:

pyenv install 3.11.9 pyenv local 3.11.9 python -m venv .venv311 && source .venv311/bin/activate python -m pip install --upgrade pip python -m pip install uagents==0.9.2 hyperon==0.2.6 uvicorn==0.22 pydantic==1.10.14
chromadb==0.5.5 sentence-transformers==3.0.1 beautifulsoup4==4.12.3 numpy==1.26.4 requests==2.32.3

Then restore agents/rag/knowledge.py (MeTTa version), and keep the same run steps.

Notes

No API keys needed for this demo.

To add an LLM (Claude) later, you’ll set ANTHROPIC_API_KEY and run a new claude_agent (see workflow below).

If you see “Agent mailbox not found,” → open the Inspector link from logs and complete Connect → Mailbox once per agent.

If you see “address already in use,” → give each agent a unique port.