google-deepmind/lab2d

A customisable 2D platform for agent-based AI research

What it solves

It provides a high-performance, easy-to-use system for creating 2D grid-world environments specifically designed for training and testing machine learning agents.

How it works

The system uses a combination of simple text-based maps to define the world layout and Lua code to define the environment's behavior. Agents interact with these worlds through either a Python API (via dm_env) or a custom C API, supporting multiple agents simultaneously.

Who it’s for

Machine learning researchers and developers who need a flexible, lightweight 2D environment to prototype and evaluate reinforcement learning agents.

Highlights

  • High performance implementation written in C++ and Lua.
  • Simple text-based map definitions for rapid world creation.
  • Support for multiple agents.
  • Compatible with the standard dm_env Python API.

Related

  • Project
    Farama-Foundation/Minigrid

    A collection of lightweight, customizable discrete grid-world environments for Reinforcement Learning research, including support for grounded language learning via BabyAI.

  • Project
    datachain-ai/datachain

    DataChain is a Python library that turns files in S3/GCS/Azure (or local) into versioned, typed datasets stored in a lightweight SQLite DB. It provides fast, sub‑second filtering, joins, aggregations and vector similarity over millions of records without moving the raw files. Pipelines are written as ordinary Python functions; the engine handles parallel execution, checkpointing and incremental updates. Optional features include an auto‑generated markdown Knowledge Base for humans/LLMs and a skill that makes the data layer visible to Claude, Cursor, Codex, Copilot or Pi. A hosted “Studio” service adds shared dataset registry, distributed compute and UI for large media. Install with `pip install datachain` and (optionally) `datachain skill install --target claude`.

  • Project
    confident-ai/deepeval

    DeepEval is an open‑source Python framework that lets you write pytest‑style tests for LLM‑powered apps (agents, chatbots, RAG pipelines). It ships >30 built‑in metrics—including LLM‑as‑a‑judge scores like G‑Eval and JevEval—as well as agent‑, RAG‑, multimodal‑, and bias‑related checks. Metrics run locally or with any LLM you choose. The library integrates with LangChain, CrewAI, OpenAI, Anthropic, AWS AgentCore, Google ADK, LlamaIndex, and even the TypeScript AI SDK, providing decorators and callbacks to capture full execution traces. A simple CLI (`deepeval test run …`) works in CI/CD, and the optional Confident AI platform adds cloud reporting and governance. Install with `pip install -U deepeval`, write `LLMTestCase` objects, pick a metric, and run—making LLM quality testing as systematic as traditional software testing.

  • Project
    HKUDS/DeepCode

    An open agentic coding system that uses multi-agent systems to transform natural language and research papers into production-ready code.

  • Project
    OpenBMB/ChatDev

    ChatDev 2.0 (DevAll) is an open‑source, zero‑code platform that lets you design and run multi‑LLM‑agent workflows via a visual web console or a tiny Python SDK. Define agents and their connections in YAML, launch from the browser or code, and use built‑in templates for data viz, 3‑D generation, game dev, research, etc. The backend is FastAPI, the frontend Vue 3, and the project ships Docker/Makefile support, extensible Python tools, and research‑grade orchestrators (Puppeteer, MacNet).