opencastor 2026.2.18.4

The Universal Runtime for Embodied AI

OpenCastor

The Universal Runtime for Embodied AI.

Stop writing boilerplate. Start building robot agents.

OpenCastor connects any AI model to any robot hardware through a single YAML config file. Swap brains (Claude, Gemini, GPT, Ollama) or bodies (Raspberry Pi, Jetson, Arduino) without changing a line of code.

Whether you have a $50 Amazon robot kit or a $50,000 industrial arm, OpenCastor makes it work.

Why OpenCastor?

Feature	What it means
Universal Adapter	Switch between Claude, Gemini, GPT, or Ollama with one config change
Hardware Agnostic	Built-in drivers for PCA9685, Dynamixel, Serial, and more
Zero Friction	Unbox to agent in under 5 minutes
Safety First	Hard-coded safety layers prevent LLM hallucinations from causing physical damage
RCAN Compliant	Built on the open RCAN Standard for interoperability
Messaging Built-in	Control your robot via WhatsApp, Telegram, Discord, or Slack

Quick Start

1. Install

Linux / macOS (one-liner):

curl -fsSL https://raw.githubusercontent.com/craigm26/OpenCastor/main/scripts/install.sh | bash

Windows 11 (PowerShell):

irm https://raw.githubusercontent.com/craigm26/OpenCastor/main/scripts/install.ps1 | iex

Manual install (any platform):

git clone https://github.com/craigm26/OpenCastor.git
cd OpenCastor
python3 -m venv venv && source venv/bin/activate
pip install -e ".[dev]"

Platform-specific notes

Platform	Package Manager	Notes
macOS	Homebrew (auto-installed)	Apple Silicon & Intel supported
Ubuntu / Debian	apt	Includes RPi auto-detection
Fedora / RHEL	dnf
Arch Linux	pacman
Alpine	apk
Raspberry Pi	apt	Enables I2C & camera automatically
Windows 11	winget / choco	Native PowerShell, no WSL needed

Installer flags: --dry-run (preview), --no-rpi (skip RPi detection), --skip-wizard

Verify installation: bash scripts/install-check.sh (or install-check.ps1 on Windows)

2. Run the Wizard

$ castor wizard

OpenCastor Setup Wizard v2026.2.17.11

Which Brain do you want to use?
[1] Anthropic Claude Opus 4.6 (Recommended)
[2] Google Gemini 2.5 Flash
[3] Google Gemini 3 Flash (Preview)
[4] OpenAI GPT-4.1
[5] Local Llama (via Ollama)

> Selection: 1

The wizard generates an RCAN config file, collects your API key, and optionally sets up messaging channels.

3. Run

castor run --config my_robot.rcan.yaml

Your robot is now online. Open http://localhost:8501 for the CastorDash web interface.

Swap Your Brain in One Line

Your robot's entire personality lives in a YAML config file powered by the RCAN Standard. Switch AI providers by editing one block:

# Option A: Anthropic Claude (Recommended)
agent:
  provider: "anthropic"
  model: "claude-opus-4-6"       # Best reasoning & safety

# Option B: Google Gemini
# agent:
#   provider: "google"
#   model: "gemini-2.5-flash"    # Stable. Also: gemini-2.5-pro, gemini-3-flash-preview

# Option C: OpenAI GPT
# agent:
#   provider: "openai"
#   model: "gpt-4.1"             # 1M context, strong vision. Also: gpt-5

# Option D: Local / Offline
# agent:
#   provider: "ollama"
#   model: "llava:13b"
#   url: "http://localhost:11434"

Architecture

[ WhatsApp / Telegram / Discord / Slack ]   <-- Messaging Channels
                    |
            [ API Gateway ]                 <-- FastAPI (castor gateway)
                    |
      [ Claude / Gemini / GPT / Ollama ]    <-- The Brain (Provider Layer)
                    |
              [ RCAN Config ]               <-- The Spinal Cord (Validation)
                    |
        [ PCA9685 / Dynamixel / GPIO ]      <-- The Nervous System (Drivers)
                    |
              [ Your Robot ]                <-- The Body

• Messaging Channels: Control your robot from WhatsApp, Telegram, Discord, or Slack.
• API Gateway: FastAPI server with REST endpoints, webhook receivers, and bearer-token auth.
• Provider Layer: Normalizes AI outputs into a standard Thought object (text + action JSON).
• RCAN Validation: Checks actions against physical constraints (speed limits, range of motion, collision).
• Driver Layer: Translates high-level intent (move_forward) into low-level signals (PWM, serial, I2C).

Supported Models (Feb 2026)

Provider	Models	Best For
Anthropic	claude-opus-4-6, claude-sonnet-4-5-20250929	Reasoning, safety, complex planning
Google	gemini-2.5-flash, gemini-2.5-pro, gemini-3-flash-preview, gemini-3-pro-preview	Video, multimodal, speed
OpenAI	gpt-4.1, gpt-4.1-mini, gpt-5	Instruction following, 1M context
Ollama	llava:13b, any local model	Privacy, offline, no API cost

Supported Hardware

Pre-made RCAN presets for popular kits, or bring your own config:

Kit	Price	Preset
Waveshare AlphaBot / JetBot	~$45	presets/waveshare_alpha.rcan.yaml
Adeept RaspTank / DarkPaw	~$55	presets/adeept_generic.rcan.yaml
SunFounder PiCar-X	~$60	presets/sunfounder_picar.rcan.yaml
Robotis Dynamixel (X-Series)	Varies	presets/dynamixel_arm.rcan.yaml
DIY (ESP32, Arduino, custom)	Any	Generate with castor wizard

The Perception-Action Loop

OpenCastor runs a continuous observe-reason-act cycle:

1. Observe -- capture camera frame + sensor telemetry
2. Reason -- send to AI model, receive structured action JSON
3. Act -- translate intent into motor commands with safety checks
4. Repeat -- configurable latency budget (default 200ms)

from castor.providers import get_provider
from castor.drivers.pca9685 import PCA9685Driver

brain = get_provider(config["agent"])
driver = PCA9685Driver(config["drivers"][0])

while True:
    frame = camera.capture()
    thought = brain.think(frame, "Sort the recycling from the trash.")
    if thought.action:
        driver.move(thought.action.get("linear", 0), thought.action.get("angular", 0))

Docker

cp .env.example .env                          # Add your API keys
cp config/example.rcan.yaml config/robot.rcan.yaml  # Or use castor wizard
docker compose up                             # Launch API + dashboard

# Or with a custom command:
docker compose run opencastor castor wizard   # Interactive setup
docker compose --profile gateway up           # Include messaging gateway

CLI Reference

Setup

castor wizard                                      # Interactive setup wizard
castor quickstart                                  # One-command: wizard + demo
castor configure --config robot.rcan.yaml          # Interactive config editor
castor install-service --config robot.rcan.yaml    # Generate systemd unit file
castor learn                                       # Interactive step-by-step tutorial

Run

castor run       --config robot.rcan.yaml          # Perception-action loop
castor run       --config robot.rcan.yaml --simulate  # No hardware
castor gateway   --config robot.rcan.yaml          # API gateway + messaging
castor dashboard                                   # Streamlit web UI
castor demo                                        # Simulated demo (no hardware/API keys)
castor shell     --config robot.rcan.yaml          # Interactive command shell
castor repl      --config robot.rcan.yaml          # Python REPL with robot objects

Diagnostics

castor doctor                                      # System health checks
castor fix                                         # Auto-fix common issues
castor status                                      # Provider/channel readiness
castor logs -f                                     # Structured colored logs
castor lint      --config robot.rcan.yaml          # Deep config validation
castor benchmark --config robot.rcan.yaml          # Performance profiling
castor test                                        # Run test suite

Hardware

castor test-hardware --config robot.rcan.yaml      # Test motors individually
castor calibrate --config robot.rcan.yaml          # Interactive calibration
castor record    --config robot.rcan.yaml          # Record a session
castor replay    session.jsonl                     # Replay a recorded session
castor watch     --gateway http://127.0.0.1:8000   # Live telemetry dashboard

Config Management

castor migrate   --config robot.rcan.yaml          # Migrate RCAN config version
castor backup                                      # Back up configs
castor restore   backup.tar.gz                     # Restore from backup
castor export    --config robot.rcan.yaml          # Export config bundle (no secrets)
castor diff      --config a.yaml --baseline b.yaml # Compare two configs
castor profile   list                              # Manage named config profiles

Safety & Compliance

castor approvals                                   # View/approve dangerous commands
castor privacy   --config robot.rcan.yaml          # Show sensor access policy
castor audit     --since 24h                       # View tamper-evident audit log
castor audit     --verify                          # Verify audit chain integrity

OpenCastor's safety kernel includes anti-subversion (prompt injection defense), work authorization for destructive actions, physical bounds enforcement (workspace/joint/force limits), tamper-evident hash-chained audit logs, and safety state telemetry at /proc/safety. See docs/safety-audit-report.md for the full architecture.

Network & Fleet

castor discover                                    # Find RCAN peers on LAN
castor fleet                                       # Multi-robot status (mDNS)
castor network   status                            # Network config & Tailscale
castor schedule  list                              # Manage scheduled tasks

Advanced

castor token     --role operator                   # Issue JWT for RCAN API
castor search    "battery low" --since 7d          # Search operational logs
castor plugins                                     # List loaded plugins
castor upgrade                                     # Self-update + health check
castor update-check                                # Check for newer versions

Contributing

OpenCastor is fully open source (Apache 2.0) and community-driven. We want your help.

Get involved:

• Discord: discord.gg/jMjA8B26Bq -- chat with maintainers and the community
• Issues: GitHub Issues -- report bugs or request features
• PRs: Fork, branch, and submit -- see CONTRIBUTING.md for guidelines
• Twitter/X: @opencastor

Areas we need help with:

• Driver Adapters: ODrive, VESC, ROS2 bridges, ESP32 serial
• AI Providers: Mistral, Grok, Cohere, local vision models
• Messaging Channels: Matrix, Signal, Google Chat
• Sim-to-Real: Gazebo / MuJoCo integration
• Tests: Unit tests, integration tests, hardware mock tests

Every contribution matters -- from fixing a typo to adding a new driver.

License

Apache 2.0. Built for the community, ready for the enterprise.

---

Built on the RCAN Spec by Continuon AI.