Octos Tutorial

Hands-on examples for building industrial robot agents with octos and dora-rs.

Examples

#	Example	What You'll Learn	Dependencies
01	Pipeline Basics	DOT pipeline engine, deterministic tool execution	dora-rs only
02	Safety Tiers	Permission gating, tier-based tool authorization	dora-rs only
03	LLM Agent	Free-form LLM reasoning, replanning on failure	dora-rs + Ollama
04	Human Gate	Operator approval gates, human-in-the-loop control	dora-rs only
05	SLAM Nav Sim	Visual MuJoCo simulation, obstacle avoidance	dora-rs + MuJoCo + Rerun

Examples 01-04 run in under 5 minutes with just pip install dora-rs pyarrow numpy.

Quick Start

# Install dependencies
pip install dora-rs pyarrow numpy

# Run the simplest example
cd 01-pipeline-basics
dora up && dora start dataflow.yaml --attach

Setup

git clone https://github.com/octos-org/octos-tutorial.git
cd octos-tutorial
pip install -r requirements.txt

For visual simulation (example 05), also install:

pip install mujoco rerun-sdk

For LLM examples (03), install Ollama and pull a model:

ollama pull qwen3:30b

Architecture

Octos is an agentic OS for robots. These tutorials use the Python SDK (octos_py/) which mirrors the Rust octos-agent crate.

┌─────────────────────────────────────┐
│           Octos Agent               │
│  ┌───────────┐  ┌────────────────┐  │
│  │  Pipeline  │  │  LLM Provider  │  │
│  │  Engine    │  │  (OpenAI/Mock) │  │
│  └─────┬─────┘  └───────┬────────┘  │
│        │    Tool Registry    │       │
│        └────────┬────────────┘       │
│          ┌──────┴──────┐             │
│          │ Safety Tier │             │
│          │   Policy    │             │
│          └──────┬──────┘             │
└─────────────────┼───────────────────┘
                  │ skill_request / skill_result
                  ▼
         ┌────────────────┐
         │  Robot Bridge  │ (dora dataflow node)
         └────────────────┘

Octos Python SDK

The octos_py/ directory contains a standalone Python implementation of the octos agent framework:

Module	Purpose
agent.py	Agent loop, tool dispatch, pipeline integration
pipeline.py	DOT graph parser, PipelineExecutor, CyclicPipelineExecutor
safety.py	SafetyTier, RobotPermissionPolicy, WorkspaceBounds
provider.py	LLM providers (OpenAI, Mock, Retry, ProviderChain)
tools.py	Tool trait, ToolRegistry, ToolResult
mission.py	Mission lifecycle, MissionExecutor, checkpoints
session.py	Session management, episode memory
skills.py	SKILL.md loader, skill injection
perception.py	Sensor aggregation

License

Apache-2.0