User repository on GitHub
Autonomous PlayStation 2 AI Agent with real-time visual telemetry HUD powered by TypeSafe Jev System One
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Ps2 AI Agent
A Python pipeline that captures the PCSX2 window, injects virtual DualShock 2 input and draws a broadcast-style telemetry HUD, with TypeSafe Jev System One as one of three swappable decision engines behind a BaseAgent interface.
- Stars
- 0
- Language
- Python
- Published by
- Community
- Author
- opaielsheikh
- Last verified
- Oct 2026
About this project
An end-to-end autonomous gaming agent architecture for PlayStation 2 emulation, equipped with low-latency window frame capture, virtual controller input injection, real-time broadcast-quality visual telemetry HUD, and modular decision-making powered by TypeSafe Jev System One. Defaults are pre-configured in config.py. To override: The pipeline features a Dual-Mode Controller Engine: In PCSX2 > Settings > Controllers > Controller Port 1, select XInput / Xbox 360 Controller.
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- Three engines implement one interface in agent_bridge.py: TypeSafeJevAgent, HeuristicVisionAgent and RandomAgent.
- Windows uses vgamepad over ViGEmBus for kernel-level gamepad input; macOS falls back to pynput key bindings.
- A mock PS2 window in test_emulator.py exercises the loop end to end without booting an ISO.
- Passing --record writes gameplay plus the live HUD straight to a 720p or 1080p MP4 file.
Watch out
No performance data is published. There are no latency figures for the Jev path, no token counts, no frame rates beyond the 30 to 60 fps async worker mentioned in the architecture diagram, and no comparison between the three agents, so the choice of engine has to be made on faith. Platform behaviour genuinely differs: Windows gets a kernel-level virtual gamepad with full analog sticks and pressure-sensitive triggers through ViGEmBus, while macOS gets discrete key bindings, which is a materially worse input path. Live emulation mode additionally requires a PS2 ISO. TYPESAFE_API_KEY and TYPESAFE_MODEL default to jev-latest in config.py, but the README gives no cost estimate for running a model call at the documented decision rate. 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.
Not for a first pass. Mode 1 runs test_emulator.py as a mock emulator window, then starts the agent against it with --window Mock PS2, which exercises capture, inference, injection and the HUD without an ISO. Real emulation is Mode 2 and needs PCSX2 booted with a game. Note the platform split: Windows uses ViGEmBus and vgamepad for kernel-level input, macOS uses pynput key bindings only.
The architecture puts three engines behind one interface: TypeSafeJevAgent, HeuristicVisionAgent and RandomAgent, selected with --agent jev, all returning a controller state, a decision label, a confidence and a latency. That structure means the Jev path is genuinely a distinct agent rather than a naming layer, but the README publishes no numbers for any of the three, so you cannot tell how much better Jev is than the heuristic or how much better than random.
Less than the branding suggests. TYPESAFE_API_KEY and TYPESAFE_MODEL are read from config.py, but HeuristicVisionAgent and RandomAgent ship in the same bridge, the mock window exists for offline verification, and the documented extension path is to subclass BaseAgent and implement act yourself for a local VLM or an RL policy. The Jev engine is the default path, not a hard dependency of the pipeline.
Not documented in numbers. capture.py does low-latency window capture with dynamic window tracking, using bettercam over DXGI on Windows and mss over Quartz on macOS, with TARGET_FPS defaulting to 30 and EMULATOR_WINDOW_TITLE defaulting to PCSX2. There are no benchmarks, no comparison between the two backends and no stated minimum requirements, so the capture cost is something you would have to measure yourself.
Three implement BaseAgent in agent_bridge.py: TypeSafeJevAgent, HeuristicVisionAgent and RandomAgent. You can subclass BaseAgent to plug in a custom VLM or reinforcement-learning agent.
Run test_emulator.py to launch a mock PS2 window, then start the agent against it with main.py --window "Mock PS2". This exercises the capture, inference, controller and HUD loop end to end.
On Windows a virtual gamepad via ViGEmBus (vgamepad) injects kernel-level signals with 360-degree analog sticks and pressure-sensitive triggers. On macOS the agent falls back to pynput key bindings, which is a materially worse input path.
More Jev resources like this
19GitHub 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-mcpPython22GitHub 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-jevPython4GitHub stars
Browser-use layer for CLI coding agents: Jev picks the operation and target element in one round trip per step, exposed over MCP with a CLI, a Python API and an HTTP server.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
brnyxx/jev-raPython
brnyxx/jev-raPython1GitHub stars
A Python search command for terminals and agents: Parallel Search seeds queries, a local Chromium fetches pages concurrently, and Jev scores the evidence to expand a breadth-first link queue. Ships readable output, JSON and an MCP server.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
komikat/psearchPython
komikat/psearchPython46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPythonFrom 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.