User repository on GitHub
- Forks
- 10
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev_Star
StarCraft II research repo combining Jev action selection with optional Astra plans across full-game macro control and SMAC-Hard micromanagement, with videos, replays, and a paper.
- Stars
- 49
- Language
- Python
- Published by
- Community
- Author
- sc2musa
- Last verified
- Oct 2026
About this project
Realtime StarCraft II macro control in five configurations, plus JEV and Astra micromanagement. 五种配置的实时星际争霸 II 宏观控制，以及 JEV 与 Astra 微操。. Read the paper / 在线阅读论文 · PDF · Video gallery / 视频展示 · Citation / 引用. Play the full winning games directly below. 22.4 fps, original speed, complete matches. macro-v2.1 / Astra + JEV · 12:47.90. earlier Astra + JEV macro · 15:04.20. C / P0 Astra + JEV / schema 7 · 00:21.43.
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- Macro module runs Protoss with 73 actions in realtime; micro covers 35 maps with Astra plans per map and Jev per-unit picks.
- Five macro configurations separate Astra constraints from advisory context and Jev from uniform random selection.
- Runs through a unified entry point with isolated virtual environments and per-module offline test suites.
- Publishes full winning videos, original replays, paper source, and per-game evidence files.
Watch out
Micro results are weak and macro wins depend on the constrained Astra-plus-Jev configuration; version and timing differences block single-variable causal claims. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No. The five-configuration table separates planner presence, plan constraints, and the action selector, and the README warns that archived version and service-timing differences prevent a single-variable causal reading. The strong macro row is Astra constrained plus Jev, not Jev alone.
Poorly: Jev alone took only a handful of wins across 35 maps and three episodes per map, with the documented versions ranging from a few wins to 7 wins against roughly a hundred losses. This is an experiment record, not a competitive bot.
The validated setup is Windows with Python 3.10 and a separately installed SC2 client matched by region and version, with separate virtual environments for macro and micro. Tests use mocked interfaces and need no game or paid models.
The videos at the top are selected winning games at original speed; complete evaluation results live separately in per-game evidence files and the paper data. There are dozens of verified replays alongside the selected clips.
The validated setup is Windows, Python 3.10, and StarCraft II 5.0.16.97563 installed for the Asia region. Run scripts/setup_environment.py macro and micro --video into separate virtual environments, set SC2PATH and TYPESAFE_API_KEY, and run scripts/install_maps.py all; Astra planning needs an authenticated native Codex CLI, overridable with --codex-path.
No. There are 140 macro and 37 micro offline tests that use mocked interfaces and do not launch StarCraft II or call paid models, and GitHub Actions runs the same two suites. Run them with unittest discover in each module against that module's own virtual environment.
Median response times were 0.391 s for JEV-only, 0.375 s for Astra-constrained, and 0.421 s for Astra-advisory macro play. The game advances during inference, and across the primary batches the run logged 19,605 Jev responses and 509 Astra responses spanning advancing frames; micro stays fixed-step with realtime set false.
More Jev resources like this
22GitHub stars
Terminal harness that checks ROS 2 robots with deterministic health tests, capability descriptions, SQLite history, and optional Jev evidence routing.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
YueBit/robodiag-harnessPython
YueBit/robodiag-harnessPython19GitHub stars
MCP server where one browser_goal call hands a URL, a goal, and code-checked assertions to a server-side loop driven by a decision model over live page elements.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
jiawei686/jev-ultrafast-mcpPython
jiawei686/jev-ultrafast-mcpPython16GitHub stars
Local Qwen3.5 decision models in 0.8B and 2B sizes with fitted temperatures, systemone-shaped server, playground, guard, MCP tools, and eval harness.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
lawrence3699/jev-stylePython
lawrence3699/jev-stylePython22GitHub stars
Independent open-source local decision model inspired by Jev, with a spark-s1 family plus smaller control models for agent gates, served locally with probabilities and policy actions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
abhishek085/open-spark-jevPython
abhishek085/open-spark-jevPython114GitHub stars
Browser agent where Jev picks each action from indexed page candidates while an LLM plans and reads, returning cited answers or schema-checked data with credential controls.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
agent-labs-dev/fastbrowsePython
agent-labs-dev/fastbrowsePython5GitHub stars
Autonomous Pokemon Red demo where Jev reads a compact text state each turn and answers typed Choice and Noul questions that deterministic code maps to PyBoy button presses.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
milanboers/jev-plays-pokemonPython
milanboers/jev-plays-pokemonPythonFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.