Are the 0.8B versus 4B accuracy numbers trustworthy?

Yes: pull both the small and larger tev1 tags, install requirements, and run the bench script; the repo keeps failing cases in on purpose and warns the sets are tiny and synthetic.

What setup mistake makes every example fail at once?

Ollama 0.35 or newer for the /v1/systemone endpoint, Python 3.10 or newer, and disk for both models; without the endpoint version the calls have nowhere to go.

How do I actually run one triage case and see what was sent?

Start the uvicorn lab, open a text case by id with an optional model flag, or open the games page; the side pane shows the real request, response JSON, and latency of every call.

Can I add my own routing or moderation case to the lab?

Append the questions plus labelled items to the scenarios file and restart the server; choice criteria stay an object of options while score criteria stay an ordered array.

What do I need to run the lab?

Ollama 0.35 or newer for the /v1/systemone endpoint, Python 3.10 or newer, and about 5 GB free disk for the two models. Pull tev1:0.8b (812 MB) and tev1 (4B, 4.5 GB), then pip install -r requirements.txt and python -m uvicorn server:app --port 8791.

What categories do the 50 use cases cover?

Triage and routing, safety, text analytics, data and search, dev tools, agent control, games, and life and fun. Everything ran on a laptop with an RTX 4060 8 GB at $0 per call.

How do I reproduce the benchmark numbers?

Run python bench.py tev1:0.8b,tev1:latest, which writes results.json. The data is small, hand-written, and synthetic, 6 to 8 labelled items per use case and 291 decisions total. Overall the 0.8B scored 211/291 at 190 ms average and the 4B 265/291 at 358 ms.

CommunityJev Demos & Experiments

Tev1 50 Use Cases

Fifty runnable tev1 decision-model examples through local Ollama, from ticket triage and guardrails to six playable games, with a live request viewer and scoreboard.

See highlights

Open github.com

Stars: 6
Language: Python
License: MIT
Published by: Community
Author: PromptEngineer48
Last verified: Oct 2026

PromptEngineer48/tev1-50-use-cases

User repository on GitHub

View on GitHub

50 use cases for tev1 (Together AI decision model) running locally via Ollama /v1/systemone: triage, moderation, routing, guardrails, agents, games.

Forks: 3
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

50 things you can do with tev1 (Together AI's open-weight decision model) running locally through Ollama: ticket triage, moderation, routing, guardrails, log triage, code review helpers, agent controls, and six games where every move is one model call. No chat, no prompts to engineer: you send some state text plus typed questions (choice, noul yes/no, score) and get an answer with probabilities back from POST /v1/systemone. Everything here ran on a laptop with an RTX 4060 8 GB, at $0 per call.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Routing Jev for Tool calling Jev for Classification Jev for Moderation

Used inJev for Software & DevTools

Languages: Python

Highlights

Runs tev1 locally through Ollama with no per-call cost, using choice, noul yes/no, and score questions.
Uvicorn lab shows live request and response JSON with latency, plus triage, safety, search, dev-tool, and agent cases.
Six games where every move is one model call, alongside lane and meter views for text cases.
Reproduce script and labelled scenarios included, with failing cases kept to show model limits.

Watch out

Scenario labels are small hand-written synthetic sets, so treat the per-use-case scores as a feel for the models rather than a benchmark. 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.

More Jev resources like this

jev-kit

Claude Code kit that guards PreToolUse calls with a free pre-filter plus one Jev question for ambiguous cases, and adds sub-agent sizing, rechecks, search, daemon speedups, and monitoring.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

jonathanavis96/jev-kitPython

robodiag-harness

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

jev-style

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

fastbrowse

114

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

macos-computer-use-kit

macOS computer-use kit exposing an 18-tool stdio MCP server plus CLI that drives apps through the accessibility tree, posting input in the background without moving the user cursor.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

Sur-Cai/macos-computer-use-kitPython

jev-ra

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

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai