What is FoodforThought?

FoodforThought is an open-source robotics data infrastructure for organizing and sharing robot learning datasets. It supports VLA (Vision-Language-Action) training, diffusion policies, and behavioral cloning with full data lineage tracking from raw demonstrations to trained models.

Is FoodforThought compatible with LeRobot?

Yes! FoodforThought natively supports LeRobot format import and export. You can import datasets directly from HuggingFace Hub and export your datasets in LeRobot-compatible format for training with ACT, Diffusion Policy, and other methods.

What robots does FoodforThought support?

FoodforThought supports a wide range of robots including humanoid platforms (Figure, Unitree H1/G1, Boston Dynamics Atlas), manipulators (Franka, UR5, WidowX, ALOHA), and mobile robots (Spot, Fetch, TurtleBot). We also support egocentric and factory data.

What training methods does FoodforThought support?

FoodforThought provides training recipe templates for popular methods including OpenVLA, Octo, RT-1/RT-X, Diffusion Policy, and LeRobot ACT. Datasets include metadata for action space, observation types, and episode statistics.

How do I contribute datasets to FoodforThought?

You can contribute datasets by signing up for a free account and using our upload page. We support multiple formats including HDF5, RLDS, ROS bags, Zarr, and LeRobot format. All contributions are credited to you.

FoodforThought CLI

Build, share, and deploy robot skills from your terminal. 30+ commands for the complete robotics development workflow.

v0.4.1

Quick Install

Get started with a single command

# Using pip (Python 3.9+)
pip install foodforthought-cli

Getting Started

Three commands to start building robot skills

Create a project

Initialize a new robot skill project with FoodforThought.

ate init my-robot-project

Generate a skill

Scaffold a manipulation skill from built-in templates.

ate generate pick-and-place

Check compatibility

See if your skill transfers to another robot type.

ate check-transfer --source ur5 --target franka

Features

Skill Generation

Scaffold robot skills in 30 seconds. 5 templates: pick & place, navigation, inspection, assembly, pouring.

Compatibility Scoring

Check if skills transfer between robot types. Kinematic, sensor, and compute scoring across 22+ embodiments.

Data Labeling

Upload demonstration videos for community labeling. Bounding boxes, action segmentation, polygon masks, Q&A.

MCP Integration

89 MCP tools for Cursor and VS Code. Build robot skills without leaving your editor.

Dataset Management

Upload, version, and export robotics datasets. Full lineage tracking from raw data to trained skill.

Hardware Discovery

Scan for robots via BLE and serial. Protocol management for UART, CAN, SPI, I2C.

Agent Memory

Persistent knowledge store for AI agents. Think, recall, and organize context across sessions with hybrid vector + lexical search.

Command Reference

All available commands at a glance

Projects

ate init <name>

Create a new robot skill project

ate clone <repo-id>

Clone an existing project

ate generate <template>

Scaffold a skill from a template

ate compile <skill-path>

Compile skill for deployment

Skills

ate check-transfer --source <a> --target <b>

Check skill compatibility between robots

ate adapt <source> <target>

Adapt skills between robot types

ate skill init <name>

Create a skill abstraction

ate skill test <file>

Test a skill (dry-run)

Data & Labeling

ate upload <video> -r <robot> -t <task>

Upload demo for labeling

ate labeling-status <job-id>

Check labeling job progress

ate pull <skill-id>

Download labeled skill data

ate data promote <artifact-id>

Promote artifact to next stage

Agent Memory

ate memory think "text"

Add a thought to the active memory

ate memory recall "query"

Search the active memory

ate memory status

Show active memory, train, and frame count

ate memory train [name]

List or switch trains of thought

ate memory new <name>

Create a new named memory

ate memory use <name>

Switch to a different memory

Marketplace

ate marketplace search <query>

Search for published skills

ate marketplace show <skill>

View skill details

ate marketplace install <skill>

Install a skill locally

Hardware

ate ble scan

Scan for BLE-enabled robots

ate ble enumerate <device>

List device characteristics

ate protocol list

Show supported hardware protocols

Auth

ate device-login

Authenticate via device flow

ate whoami

Show current user and connection

ate logout

Clear stored credentials

Robot Development

End-to-end robot skill development — from scaffolding to compatibility testing.

Project Scaffolding

Initialize and scaffold robot skill projects with templates and best practices.

Cross-Robot Transfer

Analyze skill compatibility across robot types with kinematic and sensor scoring.

Safety First

Built-in safety validation and checks (coming soon — use `ate validate` for preview).

Robotics Commands

Skill management and hardware deployment

ate init

Initialize a new project

ate clone

Clone an existing project

ate generate

Scaffold from templates

ate check-transfer

Cross-robot compatibility scoring

ate workflow validate

Validate workflow YAML

ate workflow export

Export to ROS2 launch files

ate deploy

Deploy to robot (coming soon)

ate benchmark

Performance benchmarks (coming soon)

Agent Memory

Persistent knowledge store for AI agents. Think and recall across sessions, organize context into trains of thought, and sync to the cloud.

Think & Recall

Store knowledge and retrieve it later with hybrid vector + BM25 lexical search. LLM-assisted reranking for precision.

Trains of Thought

Organize context into separate threads — like git branches for memory. Switch between them with a single command.

Security & Integrity

Ed25519 signing, AES-256-GCM encryption, and append-only hash chains for tamper-evident memory.

Quick Start

Three commands to start using agent memory

Store a thought

ate memory think "UR5 reach is 850mm, use for tabletop tasks"

Recall later

ate memory recall "UR5 reach"

Organize into threads

ate memory train planning

Memory Commands

24 commands for persistent agent memory — storage, search, sync, and security

ate memory think

Add a thought to active memory

ate memory recall

Search active memory

ate memory status

Show active memory and stats

ate memory train

List or switch trains of thought

ate memory new

Create a new named memory

ate memory use

Switch to a different memory

ate memory init

Create a raw .mv2 file

ate memory add

Add text entry to a .mv2 file

ate memory search

Search a .mv2 file by query

ate memory info

Show stats (frames, size, indexes)

ate memory export

Export frames to JSONL or Markdown

ate memory import

Bulk-import from Markdown

ate memory merge

Merge multiple .mv2 files

ate memory push

Upload memory to cloud

ate memory pull

Download memory from cloud

ate memory list

List cloud memories

ate memory keygen

Generate Ed25519 signing keypair

ate memory sign

Sign a .mv2 file

ate memory verify

Verify signature

ate memory seal

Encrypt a .mv2 file (AES-256-GCM)

ate memory unseal

Decrypt a sealed .mv2 file

ate memory capsule create

Bundle memories into portable archive

ate memory capsule ingest

Import a capsule into local memory

ate memory providers

List available embedding providers

GitHub PyPI