Awesome Jev
Projects, demos, and tools built with Jev.
Jev answers questions about data: pick an option, return a yes/no probability, or score something on a scale. It does not write text. The examples here show what people are using it for, with links to the code where available.
44 reviewed picks. Source notes and licenses. The larger discovery catalog is separate.
Quick start · Projects · Demos from X · CUA & drivers · OpenRouter winners · Contributing
Check permissions and data handling before running a project. A listing here is not a security audit.
20-second quick-start
Install typesafe-sdk and set TYPESAFE_API_KEY in your environment. This example makes one paid request:
from typesafe_sdk import Choice, TypeSafeClient
with TypeSafeClient() as client:
result = client.system_one(
state={"document": "I was charged twice."},
questions={"team": Choice(
instructions="Which team should handle this ticket?",
criteria={"billing": None, "technical": None, "other": None},
)},
)
print(result.choices["team"].choice) # e.g. billing; actual output can differ
No key yet? python3 examples/route_ticket.py prints the request without calling the API. SDK setup.
Start with a classifier, an agent, or a local alternative.
For simple rules, parsing, or arithmetic, use code. A dedicated classifier may be a better fit for a fixed task. Jev is worth trying when the choices change or the input is messy; test it on your own examples rather than relying on a demo's timing.
Browser Use / @gregpr07. Author recording, not our benchmark. Implementation · More demos and videos.
Reviewed picks
SDKs and skills
- TypeSafe JavaScript SDK - Official JavaScript and TypeScript client.
- TypeSafe Python SDK - Official Python client, with a support-ticket routing example.
- Jev Logs - Scores log entries with Jev to decide which need a closer look.
- Jevify - Agent skill for finding places to use Jev and planning comparisons with existing code.
- Official TypeSafe skill - Official skill for writing Jev requests and choosing between Noul, Choice, and Score.
- Skillbox - Self-hosted skill library with optional Jev-based recommendations.
- Tenbin - MCP server and skill for linting Jev questions, evaluating labeled examples, and choosing decision thresholds.
- wellposed - Offline linter and agent skill for Jev requests: checks missing options, broken state references, and question types.
- jevrs - Async Rust client with typed answers, derive macros, and native or WASI transports.
- Cua Driver + jev-use - Cua Driver example in Python and TypeScript: Jev chooses an action, the driver runs it, and code checks the result.
- Jev Browser (MCP, CLI, library) - Playwright browser controlled by Jev, available as an MCP server, CLI, or library.
- Jev Browser Skill - Small browser-control reference showing how to build and execute Jev action choices.
- Jev Browser Use (existing Codex connection) - Skill that uses an existing Codex browser connection. Jev clicks and scrolls; Codex types and checks the result.
- Jev Cua (Codex/Waku) - Mac automation using Cua Driver, with optional Jev recommendations for saved workflows.
- Jev Social - Read-only Instagram, TikTok, and LinkedIn research loop: Jev Choice questions pick the platform and each bounded search, read, inspect, or finish operation; SocAI executes in the user's Chrome and the app retains source-linked evidence.
- Jev Ultrafast - Jev picks browser actions and page elements; a separate model writes text.
- Jev voice browser - Voice-controlled Playwright browser. Jev chooses actions from partial speech and page controls.
- jev-bot (@compootor) - Native Mac controls through a JavaScript MCP session, with optional Jev element selection.
- TypeSafe computer use (macOS) - Mac automation using local OCR and accessibility data, with Jev choosing the next action.
- Foreman - Uses Jev to check a coding agent's progress and decide when it needs intervention.
- Jev MCP - MCP tools for checking claims against evidence, screening content, and ranking candidates.
- jev-belay - Claude Code Stop hook that checks local run evidence before Jev judges an unverified completion claim.
- jev_search - Ranks files by keyword, then checks their passages with Jev in two passes. Author marks it experimental.
- jgrep (npm: jevgrep) - Jev answers one Noul per code chunk, diff hunk, or CSV row (does this match the description?); code prints the probabilities as file:line hits, gates CI on an English rule with --diff, and lists the test files a diff can affect with --tests.
- Oko - Code-search CLI and MCP server that finds candidates locally and uses Jev to rank the source snippets.
- TypeSafe MCP - Go MCP server for making Jev requests from coding agents.
Apps and integrations
- nospace - Adds spaces while you type. Jev chooses between possible word splits; a separate model handles spelling corrections.
- tisco - Searches video transcripts with Jev, then previews clip moves and renames for approval.
- Vibe Domain - Ranks domain names by the requested style. Has a no-key heuristic mode and a hosted Jev option.
- Home Assistant Jev - Adds Jev decisions to Home Assistant entities and automations.
- llm-typesafe - Use Jev from the LLM command line or Python, with stdin, templates, and sync/async calls.
- MinusPodJev - Podcast ad-detection adapter for MinusPod: Jev scores transcript segments and code assembles the ad spans.
Games and creative projects
- Jev Chess - A shared chess game against Jev with move probabilities. Hosted demo; source code not verified.
- Jev Plays Snake - Snake where code finds legal moves and Jev picks a direction.
- jev-plays-pokemon-red - Pokemon Red agent where code builds legal actions and route state; Jev chooses only at branch points.
- JevAI for XMage - Magic: The Gathering bots that use Jev alone or alongside XMage search.
- JevPilot - Driving simulation where Jev chooses steering and speed. Not a real-world driving system.
- PROMPT FPS / Jev Doom - Jev chooses actions in a browser-based Doom engine. Uses Freedoom assets.
- TypeSafe Mario - Jev picks controller inputs from emulator state. Requires a lawful local game setup.
- Jev Music Playground - Jev picks musical parameters; code turns them into notes and MIDI. Includes an offline mode.
Independent models
Supporting drivers
All 44 reviewed picks: source notes and licenses
Computer use and drivers
Compare the browser and desktop tools. The guide separates Jev integrations from their drivers and from independent local models. Cua Driver and Browser Harness execute actions; neither is a Jev model. A locally installed tool may still send data to a hosted model.
Watch Jev in action
Original creator media, linked to the implementation. Click a preview to watch the source. Animated GIFs play inline; original GitHub video attachments appear as players below. X and YouTube open on their own platforms.
From X: demos and the original roundup
Browser Use × Jev Ultrafast
Watch a browser choose operations and DOM targets; a separate model writes text.
▶ Watch on X · Code · Source
Browser Use · @gregpr07 — Animated author recording. This narrow flight-search demonstration is not a general reliability benchmark.
|
Talk to a browser
Partial speech becomes typed intents and observed page targets; Playwright acts.
▶ Watch on X · Code · Source
Moritz Kremb · @moritzkremb — Creator post preview when accessible. Keep its browser-control server private; a recording is not a security audit.
|
Jev on the Mac desktop
OCR and accessibility observations become choices for desktop actions.
▶ Watch on X · Code · Source
@awlevin — Creator post preview when accessible. Screen content may be private; real use needs screen-recording and accessibility permissions.
|
An independent agent supervisor
Jev assesses worker progress and verification needs; code decides interventions.
▶ Watch on X · Code · Source
Josh Rosen · @JoshARosen — Architectural experiment. Not a replacement for tests, review, or permission checks.
No reliable preview image was returned. The original watch link is retained.
|
An order-book decision loop
Typed buy/sell judgments inside a bounded quoting and execution loop.
▶ Watch on X · Code · Source
Jarrod Watts · @jarrodwatts — The documented default is mock/dry-run. A dashboard does not prove live inference, profitability, or safe custody.
|
The community roundup
The discovery thread that seeded this collection; follow the original creators.
▶ Watch on X · Source
Moritz Kremb · @moritzkremb — Roundup image, not a separate working project. This directory does not claim to capture every reply.
|
OpenRouter winner: JevAI for XMage
Legal game actions become Jev choices; the hybrid combines XMage search with model judgment.
▶ Watch on X · Code · Source
ShiftSad · screenshot published by OpenRouter — Publisher screenshot, not a recording or our test. The reported 11-6-3 result belongs to the hybrid; pruning is off by default. Award announced 2026-09-21.
|
OpenRouter winner: Jev Chess
A shared chess game displays Jev probabilities over legal move candidates.
▶ Watch on X · Source
sliday · screenshot published by OpenRouter — Publisher screenshot; Source opens the project page. No public repository or code license verified. We did not play a move or reproduce the results. Award announced 2026-09-21.
|
OpenRouter winner: tisco
Search spoken content, review transcript matches, and approve clip organization in a terminal.
▶ Watch on X · Code · Source
cairodavila · screenshot published by OpenRouter — Publisher screenshot. Jev judges transcripts, not video pixels. Audio uploads, live model calls, and file changes have separate data and approval implications. Award announced 2026-09-21.
|
OpenRouter winner: Vibe Domain
Separate candidate availability checks from ranking names against a requested vibe.
▶ Watch on X · Source
OB Studio / Oliver · screenshot published by OpenRouter — Publisher screenshot; Source opens the project. The page documents a no-key heuristic mode distinct from live Jev. No public repo verified; recheck availability at a registrar. Award announced 2026-09-21.
|
OpenRouter winner: jev_search
Keyword-density code prioritizes files; Jev evaluates passages in two passes.
▶ Watch on X · Code · Source
caio0452 · screenshot published by OpenRouter — Publisher screenshot. The author labels the project AI-generated and not for production. Selected source text reaches OpenRouter; no project code was executed here. Award announced 2026-09-21.
|
Screenshots, animated demos, and implementation diagrams
Jev Tetris: see the decision
Code enumerates reachable placements; Jev chooses and the board visualizes its preferences.
Open demo / recording · Code · Source
thelau — Author UI still. A mock mode also exists; the screenshot alone does not establish live inference.
|
A structured code-review dashboard
Follow staged risk checks, evidence selection, and severity judgments in a local UI.
Open demo / recording · Code · Source
devagrawal09 — Author dashboard screenshot. Findings are review prompts, not proof of a defect or a security certification.
|
Needle: find by meaning
Score source passages and select the relevant sentence without generating a new answer.
Open demo / recording · Code · Source
Shubhamsaboo / awesome-llm-apps — Author screenshot of the semantic-find interface. Live requests use Vercel AI Gateway and send selected text.
|
How Jevthoven composes
A prompt becomes a musical plan, bounded choices, editable notes, and MIDI.
Open demo / recording · Code · Source
cocktailpeanut — Author architecture illustration, not a screenshot or an audio-model result. The full recording appears below.
|
Three primitives, made visible
Explore Noul, Choice, and Score through an interactive decision playground.
Open demo / recording · Code · Source
davila7 — Explanatory illustration, not an actual response or timing result. Hosted live examples require a key.
|
A meaning-based Mac launcher
Local search proposes files and history items; Jev ranks the intended targets.
Open demo / recording · Code · Source
dabit3 — Historical author screenshot from the archived experiment. Its README points to a successor; this is not the current UI.
|
Semantic shell suggestions
Jev ranks a bounded set of previous commands; the user accepts a suggestion.
Open demo / recording · Code · Source
mrnugget — Animated author demo uses fabricated history. Do not send sensitive real shell history without reviewing data handling.
|
A browser skill you can read
A recorded Hacker News walkthrough shows observed controls, typed choices, and execution.
Open demo / recording · Code · Source
zurfyx — Animated author recording, plus an MP4. A deliberately bounded reference implementation, not a complete browser agent.
|
Full recordings — play inline
These are the creators' original GitHub-hosted uploads, not re-encoded copies. A direct watch link is retained for clients that do not render a player.
Abide — checking project rules during an agent session — the recording could not be verified by the latest public-access check. View the creator source.
Jevthoven — from a prompt to editable music
Watch typed musical decisions turn into an editable multitrack composition.
https://github.com/user-attachments/assets/176c69e4-501e-4b71-8517-957cc692882a
Open recording · Repository · Creator source
cocktailpeanut — Code renders notes and audio. Fixture/manual modes are distinct from billable live Jev decisions.
jev-align — improve a decision with human labels
Find uncertain examples, label them, and review proposed AI-function improvements.
https://github.com/user-attachments/assets/81650587-e3f1-4655-8213-ed5f6e120e9a
Open recording · Repository · Creator source
Sutro — Experimental active-learning CLI. Training improvement is not held-out quality; inference and reflection models may both bill.
Longer walkthroughs on YouTube
A full Jev walkthrough
A longer introduction and practical demonstrations from Moritz.
▶ Watch on YouTube · Source
Moritz | AI Systems — Linked video, not an independently reproduced tutorial. Title and creator metadata are checked through YouTube oEmbed when accessible.
|
An email-classification walkthrough
A creator walkthrough of applying Jev to email classification.
▶ Watch on YouTube · Source
vogel — Linked video, not an independently reproduced tutorial. Title and creator metadata are checked through YouTube oEmbed when accessible.
|
Practical ways to use Jev
An overview of possible applications and implementation ideas.
▶ Watch on YouTube · Source
Mark Kashef — Linked video, not an independently reproduced tutorial. Title and creator metadata are checked through YouTube oEmbed when accessible.
|
Source notes: media remain hosted by their creators/platforms; this repository does not relicense them. Public X mirror metadata is used only to locate a poster, not to certify a demo. Reachable media is not proof of live inference, security, or benchmark performance. No project was executed for this gallery.
Media sources and link-check receipts · Suggest a visual demo
OpenRouter's September 21 roundup features JevAI for XMage, Jev Chess, tisco, Vibe Domain, and jev_search. Their screenshots are in the gallery above. Project links and implementation notes. These are OpenRouter's picks, not a ranking by this list.
All curated X links
Curated X demos
Moritz's roundup started this collection. OpenRouter's thread added five more projects. Full index and source notes.
| Demo |
What Jev does |
Watch / inspect |
| Voice-controlled browser — @moritzkremb |
Maps partial speech transcripts to typed intents and observed page targets; Playwright executes the selected action. |
X demo · repo · source |
| Browser Use × Jev Ultrafast — @gregpr07 |
Chooses an operation and compatible DOM target in one decision cycle; a separate model supplies typing text. |
X demo · repo · source |
| macOS computer use — @awlevin |
Selects desktop actions from OCR and accessibility-derived state, with a separate writer for free text. |
X demo · repo · source |
| Foreman coding-agent supervisor — @JoshARosen |
Assesses coding-worker progress and verification needs while deterministic policy decides interventions. |
X demo · repo · source |
| Jev trading-loop demonstration — @jarrodwatts |
Uses typed buy/sell judgments in an order-book loop, while code owns quoting, limits, and execution. |
X demo · repo · source |
| JevAI for XMage — ShiftSad · featured by @OpenRouter |
XMage supplies legal plays; pure and hybrid players use Jev for bounded decisions and selection among searched lines. |
X demo · repo · source |
| Jev Chess — sliday · featured by @OpenRouter |
A shared internet-versus-Jev chess game visualizes probabilities over legal move candidates. |
X demo · project · source |
| tisco — cairodavila · featured by @OpenRouter |
Jev searches transcript meaning; code previews and applies approved clip moves and renames. |
X demo · repo · source |
| Vibe Domain — OB Studio / Oliver · featured by @OpenRouter |
Ranks domain candidates against vibe and keyword preferences, separately from registry and availability checks. |
X demo · project · source |
| jev_search — caio0452 · featured by @OpenRouter |
Keyword-density code prioritizes files and chunks; Jev judges passages in a fast first pass and a broader second pass. |
X demo · repo · source |
Find a project
The full catalog includes community listings and automatic discoveries that have not been selected for this README. Use GitHub's outline or Cmd/Ctrl+F to browse it. For filters, clone the repo and open docs/catalog.html locally. Viewer instructions.
For more demos and skills, browse Ship with Jev. We use it for discovery, then check the original projects. Our selections from it.
Reviewed means the primary source was inspected. Indexed means it came from another directory. Auto-discovered means a README keyword match. These labels do not imply the project was tested here.
Refresh status
Last discovery: 2026-09-23T13:44:11Z (UTC). Latest repository metadata date: 2026-09-23. Unavailable repositories at last check: 14. Coverage, failures and run receipts.
The four-hour job updates the discovery catalog and checks a rotating set of links. It does not choose projects for the README. Check history and coverage.
Contribute
Suggest a project with a source link, what Jev does, and one reason it is useful. Small projects are welcome. A clear implementation or inspectable demo matters more than stars; screenshots and performance claims alone are not enough.
Found a broken link or a misleading description? Open an issue. Contribution guide · Selection and removal policy.
Safety, licensing, and evidence limits
Some projects use paid APIs, logged-in browser sessions, or desktop permissions. Read their setup and privacy notes before connecting them. Reported benchmarks belong to their authors unless a separate reproduction is linked. Mock modes and independent models are labeled where known.
This is an unofficial list. Each linked project and media file keeps its own license. A missing license is not permission to reuse. Security policy · Sources and credits.
A curated awesome-list of demos, projects, SDKs and skills built with Jev — 44 reviewed picks with source links, plus a curated gallery of demos from X and YouTube.