Jev directory
CommunityJev Agents & MCP
Jev Leveling Agent
A guide-directed leveling agent for a private TBC 2.4.3 server: ten clients on one PC follow generated quest guides with scripted, self-verifying skills, and a vision tutor takes over only when a routine fails.
- Stars
- 0
- Reading time
- 3 min
- Published by
- Community
- Author
- chalkychalk42
- Last verified
- Oct 2026
Used inJev for Gaming
- Languages
- Python
- Capabilities
- Self-hostedRuns locally
Highlights
- Most play decisions run as local scripted rules, so the agent makes progress without any model call.
- A five-tier loop — Body, Tracker, Policy, Choices, Tutor — sends only failed objectives to the subscription vision tutor.
- Alternatives with real stakes, such as hunt stations and heal timing, are learned with Thompson sampling and pooled across clients.
- 48-hour unattended runs in September took a paladin from level 13 to 15.87 and a mage from level 1 to 15.24.
- Skills verify their own effects from what the addon paints; the core contracts are testable with no client, capture, or network.
Quickstart
bash
uv venv --python 3.12 && uv pip install -e ".[dev]"
.venv/bin/python -m pytest # no game requiredWatch out
Copy reviewed against the repo README on 2026-10-15. No licence is declared and no benchmarks are published; the measured runs happened on the author's private TBC 2.4.3 server, and the private runtime assets are not in the repository. The repository returned 404 on 2026-10-16, so the captured README is now the only record of the review.
Questions about this entry
Short answers drawn from the linked project or page.
It levels characters unattended on a private World of Warcraft The Burning Crusade 2.4.3 server. The guide supplies the objective, scripted routines play it, and learned choices decide the real alternatives, such as which hunt station to use or when to heal.
Rarely, as the tutor tier. A vision tutor takes an objective only after its scripted routine has failed, so most progress needs no model call and the cost stays low as the character levels.
Every skill proves its effect by what the addon paints, rather than assuming the action landed. The wire format, state contract, situation key, verifier, and reward function are all testable with no client, no capture, and no network.
Not fully. The README says a source clone alone is not a runnable installation: private runtime assets, server connectivity, and isolated state are covered by the separate partner setup and handoff guides, and the daily routine runs against a private runtime repository.
A live client played forty-eight hours unattended in September: a paladin went from level 13 to 15.87, then a mage the campaign tool made itself went from level 1 to 15.24. Since 12 October ten clients play at once, five paladins and five mages.
Choices between real alternatives are learned from outcomes with Thompson sampling, using a danger map and route memory, and the learning is pooled across the clients, so one client's experience informs the others' options per visit.
The architecture contract, operating instructions, a decisions log with reasons and evidence, measured status, roadmap with acceptance gates, and the original plan. The README deliberately omits a client count because it goes stale quickly; STATUS.md carries the current one.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev picks which Apple HIG guides go into the agent's context. NatashaTheRobot built a small AskHIG CLI and…
X: Three decision models, one game of Snake
X: Diogo Almeida with a16z on Jev
X: Luna plans, Jev plays poker
More Jev resources like this
A local Codex and Waku plugin for app-agnostic Mac computer use: Accessibility actions with a visual fallback, approval-gated mutations, and a Jev fast path for repetitive tasks.
Jev Agents & MCP#community#catalog-import#madewithjev
Community
eronmmer/jev-cuaTypeScript
eronmmer/jev-cuaTypeScriptMakes NPCs in games talk back: skills, a self-hosted Fish Audio voice server and tools that let a coding agent wire up in-character dialogue, with TypeSafe Jev picking each NPC action and a full GTA V example.
Jev Agents & MCP#community#github#python
Community
christos-antono/NPCTalkPython
christos-antono/NPCTalkPythonBrowser-only Laya demo suite: four Vite apps — a Traditional Chinese decision playground, Gomoku, Big Two and a 3D maze — run the Laya ONNX model on-device through Transformers.js, downloading fp16 weights from Hugging Face on first use.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
killkli/open-jev-layaPython
killkli/open-jev-layaPythonAutonomous gaming agent pairing Laya's local sub-30 ms System One inference with TypeSafe Jev Choice, Score and Noul decisions. Plays Windows PC games with hardware scan codes and Android games through ADB touch injection.
Jev Agents & MCP#community#catalog-import#logicrw
Community
newuser7171/jev-gamepilotPython
newuser7171/jev-gamepilotPythonAn experimental Factorio agent in which Jev chooses goals and actions as typed questions while deterministic code owns the game rules and actuation; an offline mock loop runs with no API key and no game copy, with live and hierarchical modes documented.
Jev Agents & MCP#community#github#python
A PPO-trained reinforcement-learning decision brain for the CONNECT TYCOON company simulator from CONNECT AI LAB: it runs staffing, inventory, marketing, refund and security judgments and answers in the JEV request format with probabilities.
Jev Models & Reimplementations#community#open-models#reinforcement-learning
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.