Agentic Dora — Proof of Concept

This repository contains a local Proof of Concept (PoC) for the Google Summer of Code 2026 - Project #10 (Agentic Dora).

The goal of this PoC is to prove the core architecture: establishing a zero-copy dataflow bridge between a local LLM agent and a simulated robot node using the dora-rs framework.

Architecture

• Agent Framework: llama3.2:1b running locally via Ollama.
• Middleware: dora-rs using zero-copy Apache Arrow (PyArrow) serialization.
• Nodes:
  • agent_node.py: Queries the local LLM, forces strict JSON output for tool calling (e.g., move_to, read_cam), and publishes the command as an Arrow array.
  • robot_node.py: A mock actuator node that receives the Arrow array instantly and parses the executable JSON command.

Quick Start

1. Install Dependencies

python3 -m venv venv
source venv/bin/activate
pip install dora-rs pyarrow

2. Start Local LLM
Ensure you have Ollama installed and the model downloaded.
Bash

ollama run llama3.2:1b

3. Run the Dataflow
Start the Dora coordinator and launch the graph:
Bash

dora up
dora start dataflow.yml --attach

Expected Output

When running, the pipeline successfully demonstrates the Agent reasoning about a natural language prompt, generating a structured JSON action, and the Robot node successfully receiving it over the Dora dataflow in real-time.