AIエージェント
Qwen-AgentWorld
Qwen-AgentWorld: a language world model simulating MCP, terminal, web, and OS environments for agents.
★ 1.0k スターApache-2.0
ガイドを見る →agent-world-modelは、Snowflake-Labsが公開しているオープンソースのAIエージェントプロジェクトです（GitHub: Snowflake-Labs/agent-world-model）。GitHubスターは465、フォークは50で、NOASSERTIONライセンスの下で公開されています。GitHubでの最終更新は2026年5月28日です。
Snowflake's Agent World Model: synthesise 1,000 executable MCP tool-use environments for agentic RL.
Decide whether to use the 1,000 released environments directly or synthesise your own from seed scenarios.
Clone the repository, run uv sync, and download the released environments dataset from Hugging Face if you are using them.
Set your LLM provider credentials and model choices for synthesis and verification.
Start an environment's MCP server, run the demo agent against a task, and verify its trajectory with the built-in judges.
The awm CLI generates scenarios, tasks, database schemas and sample data, API specifications, MCP environment code, and verifiers, or runs the whole pipeline at once.
The synthesised environments, tasks, databases, and verification are published on Hugging Face and can also be tried in a live demo Space.
Every environment is exposed as an MCP server, so agents interact through the same list_tools and call_tool protocol.
Start servers per scenario, health-check one or all environments, and reset databases to their initial state between runs.
Verify agent trajectories with a code-augmented LLM-as-a-judge or a purely code-based judge.
An adapted evaluation harness covers BFCLv3, tau2-bench, and MCP-Universe under the same MCP tool protocol, with trained Arctic-AWM models released alongside.
The project uses uv for its Python environment.
git clone https://github.com/Snowflake-Labs/agent-world-model.git
cd agent-world-model
uv syncGrab the 1K environments dataset and set the synthesis LLM provider variables.
hf download Snowflake/AgentWorldModel-1K --repo-type dataset --local-dir ./outputs/
export AWM_SYN_LLM_PROVIDER="openai"
export OPENAI_API_KEY=<your key>Serve a model with vLLM, start a scenario's MCP server, then run a task.
awm env start --scenario "scenario_name" --envs_load_path outputs/gen_envs.jsonl --port 8001
awm agent --task "show me the top 10 most expensive products" --mcp_url http://localhost:8001/mcp --api_url http://localhost:8000/v1 --model Snowflake/Arctic-AWM-4BIt is developed by Snowflake AI Research with UNC-Chapel Hill collaborators, under the Snowflake-Labs organisation; the README notes its infrastructure was merged into Meta PyTorch's OpenEnv.
The code and the 1,000 environments are openly released on GitHub and Hugging Face, but the catalog metadata lists no license, so check the repository's license file before relying on specific terms.
For synthesis and LLM-judge verification, yes: you configure an OpenAI or Azure OpenAI key. Using the released environments with a locally served model needs no key.
Hand-built environments like a single benchmark suite are fixed; AWM generates environments itself, so you can scale to a thousand worlds and synthesise new ones for new domains.
Clone the repo, uv sync, download Snowflake/AgentWorldModel-1K from Hugging Face, start an environment with awm env start, and run the demo agent.
Qwen-AgentWorld: a language world model simulating MCP, terminal, web, and OS environments for agents.
Jev Ultrafast — the fastest browser agent from browser-use: one decision-model request per action cycle drives a real browser, MIT
ESP32 iPhone Agent — let an AI model use a real iPhone with a $7 ESP32: AirPlay eyes, USB HID hands, no computer, no jailbreak
Open-source Computer Use as an MCP service: let any agent drive macOS, Linux, and Windows desktops.
An Open Phone Agent Model & Framework. Unlocking the AI Phone for Everyone
Tongyi Deep Research - Alibaba's open-source deep research agent and 30B model
LibreAIカタログに掲載されたAIエージェントプロジェクトの中で最もスターの多いもの — agent-world-modelが合わなかった場合に検討したい、現実的な候補リストです。
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
カタログの中でagent-world-modelとスター数が近いAIエージェントプロジェクト — 前後の棚もそのまま見て回れます。
Guardrails for building secure, robust agents.
A harness-oriented framework for production-grade agent applications.
Detect prompt injections and jailbreaks before they hit your LLM.
The tougher, more robust successor to the MMLU benchmark.
A benchmark testing whether agents can solve long-horizon coding tasks.
Safety testing for agents, right in pytest.