MushyBook
brain2qwerty logo

brain2qwerty

Non-invasive decoding of typed sentences from MEG and EEG brain recordings using a convolutional encoder, transformer, and character-level language model.

Developer ToolsPythonNOASSERTION 851 125Health 85
GitHub

What is it?

What it is

A research project and codebase designed for decoding typed sentences from non-invasive recordings of the human brain.

Why it exists

To enable the reconstruction of natural language sentences from brain activity captured via non-invasive methods, such as MEG scanning.

Who should use it

Teams building with Python who want an open-source, self-hosted option.

Who should avoid it

Teams that need a fully managed SaaS with enterprise SLAs out of the box.

How it works

A quick walkthrough in plain English

How brain2qwerty works

Step 1 of 3

You interact with it

Open brain2qwerty, send a request, or connect it to your stack.

Features

Non-invasive decoding of typed sentences from human brain activity using MEG recordings
Python-based implementation with modular architecture (brain2qwerty_v1 and v2)
Integration with NeuralSet and NeuralTrain for data management and training pipelines
Open-source code released under CC BY-NC 4.0 license
Supports decoding of natural language sentences with high accuracy
Provides a project website and detailed documentation
Includes preprint and peer-reviewed publication references
Designed for research and potential future real‑time applications
Utilizes publicly available Spanish dataset (BCBL) for training and evaluation

Advantages

  • Open source (NOASSERTION)
  • Active Python ecosystem
  • Self-hosted deployment options

Disadvantages

  • Requires operational ownership for self-hosted setups
  • Community support varies by project maturity

Installation

native

git clone <repo-url>
cd <repo>
# See README for language-specific setup

FAQ

What is Brain2Qwerty?

Brain2Qwerty is a tool that decodes sentences from non-invasive brain recordings, such as MEG data, by translating neural activity into typed text.

What license does Brain2Qwerty use?

The code is released under the CC BY-NC 4.0 license, which allows non-commercial use with proper attribution.

How can I access the datasets?

Brain2Qwerty v1 datasets are available on Hugging Face (https://huggingface.co/datasets/bcbl190626/SpanishBCBL), while v2 datasets are under embargo until the paper's acceptance.

Where can I find the publications?

The key publications are available at https://www.nature.com/articles/s41593-026-02303-2 (Nature Neuroscience, 2026) and the preprint link in the README.

What are the differences between v1 and v2?

v2 is an updated version under embargo until the paper's acceptance, while v1 is publicly available and includes earlier implementations.

What frameworks or tools does Brain2Qwerty rely on?

It uses NeuralSet and NeuralTrain, open-source frameworks developed by Facebook AI Research for neural data processing and training.

Loading documentation…
View on GitHub

Featured in Videos

YouTube tutorials and walkthroughs for brain2qwerty

t5

70,223,459 views

Starts at 00:00

Share

Alternatives

Similar projects ranked by category, topics, and text overlap.

Compare
muscriptor logo

muscriptor

Python

MuScriptor offers three transformer decoder-only variants (small, medium, large) trained on 170k songs, accessible via HuggingFace with a CC BY-NC 4.0 license. It provides a Python API, CLI, and a FastAPI-powered web UI for real-time transcription, piano-roll visualization, and MIDI export, with support for GPU/CPU, instrument filtering, batching, and beam search.

824·Health 77

Compare
DeepSpec logo

DeepSpec

Python

DeepSpec consists of three stages: data preparation (downloading prompts, regenerating target answers, building a target cache), training (launching train.py with configurable algorithms and target models via config files, multi-GPU support), and evaluation (running eval.py on benchmarks such as GSM8K, MATH500, HumanEval, etc.). It includes released checkpoints for Qwen3 and Gemma families and integrates code from SpecForge, DFlash, Qwen3, and Gemma.

6.7k·Health 90

Compare
lift logo

lift

Python

A 9B vision model (lift) for extracting structured JSON from PDFs and images using schema-constrained decoding. It provides CLI, Python API, Schema Studio app, and supports local (HuggingFace) and remote (vLLM) inference. Benchmarks show 90.2% field accuracy and 9.5s median latency. Open source (Apache 2.0) with model weights available under OpenRAIL-M for non-competitive use.

859·Health 82

Compare
agent-as-a-router logo

agent-as-a-router

TypeScript

ACRouter provides a system for routing coding tasks to various backend models, balancing accuracy and cost. The repository includes the CodeRouterBench dataset, reproducible implementations for in-distribution and out-of-distribution (OOD176) testing, and runtime integrations for tools like Claude Code and Codex. It supports both offline reproduction using cached results and live inference-time integration with verifiers.

881·Health 77

Compare
deepsec logo

deepsec

TypeScript

deepsec follows a multi‑stage pipeline: a fast regex‑based scan identifies candidate files, AI agents (via Vercel AI Gateway or local Claude/Codex) investigate those files to emit findings and recommendations, and optional revalidation checks git history for fixes. The tool supports distributed execution across Vercel Sandbox microVMs for large monorepos, provides a lightweight triage model, and includes plugins for enrichment. Security is mitigated by sandboxing, limiting API key exposure and network egress. Users bootstrap with `deepsec init`, run scans with `pnpm deepsec scan`, process results, and export findings in Markdown or JSON.

5.4k·Health 90

Compare
brain2qwerty | MushyBook