AIRI Launches Open Community for AI Agents in NetHack
Researchers from AIRI have launched an open community to solve NetHack using AI planning systems, featuring evolutionary harnesses and independent verification.
Researchers from the AIRI lab have announced the launch of an open community dedicated to solving the complex strategy game NetHack using artificial intelligence planning systems.
The initiative invites everyone working on autonomous agents and strategies. A key feature of the format is that each participant can have their own way of obtaining a program, their own tools, and their own strategy. The project's shared infrastructure allows participants to see others' results, compare them, and continue developing solutions based on colleagues' work.
For the start, an evolutionary harness based on the MAP-Elites algorithm is available. Participants can use tools like Claude Code, Codex, or OpenCode to modify bot code. The algorithm automatically saves the best solution variants for different starting game characters.
A fundamental distinction: inside the game, the created computer program acts, not a Large Language Model (LLM) on every turn. Participants can develop their own harnesses, and results are independently re-verified on hidden seeds to ensure fair competition.
Try to solve NetHack here: AIRI NetHack Challenge
Why it matters
AnalysisThis challenge demonstrates the shift from simple LLM usage to creating autonomous agents that generate their own code to solve complex, long-horizon tasks in partially observable environments.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Hands-on Guide: Autonomous AI Agents on Open-Source Stack & Free APIs
If you want to build an AI agent, the first question is usually not \"which architecture to choose\", but \"where do I even start and what do I need\".
OpenAI launches retro console for vibe coders
OpenAI has released ModRetro Chromatic Codex Edition, a $200 Game Boy-compatible console with a rewritable cartridge and a Codex plugin for building games.
Claude Code Projects: How a Single AI Orchestrates a Team of Autonomous Agents
A deep dive into Claude Code Projects: Coordinator and Threads architecture, parallel branches, shared memory via MEMORY.md, and cloud agent management.
Agent Observability & Tracing (OpenTelemetry)
Methods for collecting metrics, tracing reasoning chains (Spans), analyzing tool latencies, and monitoring token costs using OpenTelemetry and specialized platforms (Langfuse, Arize).