Playground/Home
API ACTIVITY

Usage & budget

PLAYGROUND SAFETY LIMITS

Steady requests. Visible limits.

0 active · 0/20 queued

Browser: one start every 1.2 seconds (at most 50/minute), up to 2 active calls and 20 waiting. Shared by examples in this tab. Waiting requests cancel when their run is stopped; a full queue rejects new work without an API call.

Server: 60 requests per rolling 60 seconds and 2 concurrent calls per key, per server instance. Other tabs using the same key share that instance’s allowance. This in-memory safeguard resets on restart and is not a distributed or account-wide quota.

Source: playground policy in lib/requestRateLimit.ts. TypeSafe may impose separate limits. Provider HTTP 429 pauses calls until its reported reset; without a reset time, retry requires your action. HTTP 402 is a budget/billing failure, not a pacing limit. No automatic retries or background replay.

Your TypeSafe API key

Import your key to override the shared server default.

Saved on this browser until removed or site data is cleared. Masked on screen; sent only with Jev requests through this app. The saved value is never displayed.

Connecting

TYPESAFE AI / COMMUNITY PLAYGROUND

Small model.
Many possibilities.

Explore Jev through 19 hands-on examples. Give it context, define the choices, and watch a decision take shape.

WORKSPACE GUIDE

Home

Find an experiment, understand its decision, and inspect the evidence.

Pick a taskTry an exampleInspect evidence

What you provide

A task or question you want to explore. Search matches workspace names, descriptions, and groups.

What happens

The directory filters the available workspaces locally. Opening a card takes you to a dedicated experiment with its own controls and evidence.

What you get

A starting point for language, workflow, code, or simulation tasks. Browsing this directory makes no model call.

Walk through a run

  1. 01

    Choose a question

    Browse the four workspace groups, or search by the task you want to explore.

  2. 02

    Open an example

    Each card shows the input and the kind of result. Start with a preset before changing the context.

  3. 03

    Inspect the outcome

    Use the workspace guide to understand the decision, its evidence, and its execution limits.

Try this

Search for robot, open MicroDuck, and compare its seeded controller with a live Jev run when you are ready.

Execution limits

This is an independent community playground. Live model outputs, seeded demos, and solver checks are labeled separately; none proves general model quality.

Explore the example. Check the evidence.
ContextFixed choicesDecision

A classifier at the center.
You control what happens next.

19 of 19 examples

Language & data7

Turn unstructured context into a clear, grounded choice.

Example builder

Explore the original catalog, edit inputs and define your own closed-set questions.

in
Context
out
typed answer

Jev Chat

Explore Jev-driven response composition, source passages, constrained fiction, and visible decisions.

in
Question
out
composed response

Conversation lab

Rank messages and find the person your bot should answer next.

in
Messages
out
best reply target

Document extraction

Select dates, vendors and amounts from values actually present in a document.

in
Document
out
selected fields

YouTube extract

Select original caption passages with Jev scores, cost tracking and timestamp-linked verification.

in
Video captions
out
chronological extract

Vector reranker

Compare Jev, vector similarity and a lexical baseline on the same candidates.

in
Search results
out
relevance order

Meme lab

Classify a meme’s humor, audience and clarity from reviewed captions.

in
Caption
out
audience fit

Agents & workflows6

Choose the next step, with explicit policy and approval gates.

Ask gate

Decide whether a request belongs with Jev or needs human attention.

in
Request
out
reviewer

Workflow chat

Describe a scenario and see which action follows your process.

in
Scenario
out
next action

Tool router

Follow a request through a graph with blocked tools and approval stops.

in
Request
out
permitted tool

LangChain

Try a typed routing tool with live classification or a mocked response.

in
Tool call
out
governed route

Clean-room rebuild

Rebuild observed screens and API contracts, then compare behavior in a real browser.

in
Observed app
out
verified rebuild

Browser agent

Drive a synthetic flight search from an indexed element table, one operation and target per request.

in
Element table
out
operation + target

Code & governance3

Surface risky changes and verify decisions against fixed rules.

PR review

Paste a PR or diff to separate routine changes from hunks needing review.

in
Diff
out
review gate

AST governance

Trace changed symbols, affected callers and deterministic policy findings.

in
Symbols
out
change impact

SMT solver

Compare Jev’s prediction with an exact solver on structured constraints.

in
Constraints
out
verified result

Games & simulations3

Watch small decisions play out—and discover their limits.

Jev plays Doom

Enter JevDoom’s fullscreen 3D arena and compare human, Jev and random play.

in
Arena state
out
one action

MicroDuck arena

Guide a robot through a 3D delivery floor with seven possible actions.

in
Robot sensors
out
movement

Jev attempts chess

See why choosing a legal move without lookahead falls short of a chess engine.

in
Legal moves
out
one choice