User repository on GitHub
Like Claude Plays Pokemon, but with Jev
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev Plays Pokemon — zbloss
A Pokemon Red agent that boots a ROM, picks each move with Jev, and streams video plus a live decision panel from a small read-only local HTTP server.
- Stars
- 1
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- zbloss
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Plays Pokemon — milanboers.
About this project
Like Claude Plays Pokemon, but with Jev. Copy .env.example to .env and fill in at least TYPESAFEAPIKEY, then: Every .env value has a matching global CLI flag (e.g. --openai-api-key, --typesafe-api-key) that overrides it for a single invocation - see uv run jev-plays-pokemon --help and docs/adr/0001-unified-typer-cli-and-config-settings.md. run starts a small read-only HTTP server on --stream-port (default 8080) and logs its viewer URL at startup. It serves three endpoints:
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- run, benchmark, watchdog and a ROM-path variant, each with a global CLI flag that overrides the matching .env value.
- run serves a combined viewer page, a bare /video.mjpg feed, and a raw JSON endpoint meant for polling, not browsers.
- benchmark takes explicit --turns and --warmup counts so a measured run has a defined length and warm-up.
- watchdog runs the player as a monitored subprocess that restarts it automatically.
Watch out
This is the thinnest README in the set, and it only documents usage, the three endpoints and one ADR reference. Not documented: which emulator or ROM loader is used, what the model is shown per frame, the prompt or question wording, any benchmark result, token usage or cost, the test setup, or the license. The JSON endpoint's purpose is explicit - it is meant for programmatic polling, not for opening in a browser. The CLI also exposes flags such as --openai-api-key alongside --typesafe-api-key, so more than one provider is configurable, but the README says nothing about what the non-TypeSafe keys are for. Setup requires copying .env.example to .env, filling in at least TYPESAFE_API_KEY, and supplying your own ROM through --stream-port and --rom-path. The README we reviewed does not describe the model, the emulator or any results, so treat this as a starting point to read the code rather than an evaluated integration. 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.
The README does not say. It documents the CLI, the endpoints and that the decision payload carries an action, a confidence and milestones, but not the emulator, how the frame reaches the model, or the question wording. The detailed design decisions are deferred to an ADR in docs/adr, which is where you would have to read next.
Yes. run starts a small read-only HTTP server on --stream-port, default 8080, and logs its viewer URL at startup; /viewer is the combined video plus live decision sidebar page and is the one to open in a browser or use as an OBS browser source. /video.mjpg is the bare MJPEG feed for setups that want video and decision data as separate sources, and the root endpoint returns the latest decision as raw JSON for programmatic polling.
Your own ROM file, passed as --rom-path pokemon_red.gb; none is bundled. Copy .env.example to .env and fill in at least TYPESAFE_API_KEY first, and run serves a small read-only HTTP server on --stream-port, default 8080. The README does not say which emulator or ROM loader it drives, so no particular dump is documented, and while flags such as --openai-api-key sit next to --typesafe-api-key, it never says what those other keys are for.
No. The benchmark subcommand exists and takes --turns and --warmup counts, but the README publishes no benchmark numbers, no token cost and no comparison against a scripted baseline. It also documents no test setup or license, so you would be evaluating the code yourself.
run boots the ROM and plays it, benchmark takes explicit --turns and --warmup counts so a measured run has a defined length and warm-up, and watchdog runs the player as a monitored, auto-restarting subprocess. All are subcommands of the same uv entry point, so a typical measured invocation is uv run jev-plays-pokemon benchmark --turns 25 --warmup 5.
No. Every .env value has a matching global CLI flag, such as --openai-api-key or --typesafe-api-key, that overrides it for a single invocation only. You still copy .env.example to .env and fill in at least TYPESAFE_API_KEY first, but the flags let you change one setting without editing the file; run uv run jev-plays-pokemon --help for the full list.
By design, not a bug. run serves three endpoints: /viewer is the combined video plus live decision sidebar page to open in a browser or use as an OBS browser source, /video.mjpg is the bare MJPEG feed for setups that want video and decision data as separate sources, and / returns the latest decision as a raw JSON snapshot of action, confidence and milestones for programmatic polling.
More Jev resources like this
62GitHub stars
Simulation study that drives one xArm7 apple-to-plate task with Jev, a large reasoning model, and a small model through identical incremental controls.
Jev Demos & Experiments#community#jevbest#catalog-import
10GitHub stars
Agent bridge that connects any CDP-speaking browser to Jev by reading the DOM instead of the rendered layout, so headless or non-drawing browsers can still drive CLICK, TYPE_TEXT, SELECT, WAIT, DONE, and BLOCKED steps.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
lexmount/jev-browser-bridgePython
lexmount/jev-browser-bridgePython114GitHub 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/fastbrowsePython4GitHub 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-raPython2GitHub stars
A Python decision layer for browser and desktop automation: your code observes the surface, builds the candidate list and verifies the result, and the model only picks one id per step.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
aidil2105/jev-browser-pilotPython
aidil2105/jev-browser-pilotPython0GitHub stars
A local Python browser automation UI that turns a URL and a plain-language goal into bounded actions, with Jev picking the next operation from observed candidates and a separate text model used only to fill fields.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
KesavanKing/jev-browserPython
KesavanKing/jev-browserPythonFrom 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.