close
typesafeai.appUnofficial community project

Independent capability index · 2026

What can Jev actually do?

Jev is TypeSafe's System One model. It does not chat. It returns typed decisions your code can act on. This index tracks 151 public projects that demonstrate it — each one checked against its artifact, each one carrying its own limitations.

Browse all 151 recordsSubmit public evidence

Unofficial and independent. Every published capability links to evidence, shows how it was verified, and identifies whether the example came from TypeSafe or a community publisher. Last full re-check: 2026-10-09.

How the index works

How a record earns its place

Anyone can claim a Jev demo. This index only publishes what can be checked, and says exactly how far the checking went. Four rules govern every record:

  1. 01

    It starts from a public artifact

    A repository, a published demo, a cookbook example, or a public post — something a reader can open and inspect. Submissions without a checkable artifact are rejected, and every record links back to its source with the date we captured it.

  2. 02

    The evidence level is explicit

    E1 means the author reported it and we could not yet verify the artifact. E2 means the artifact itself was inspected and matched the claim. E3 means we reproduced it ourselves. The level sits on every card and page — no record outruns its evidence.

  3. 03

    Limitations are mandatory

    Every record must list what its evidence does not prove — single runs, one browser profile, no independent audit, whatever applies. Reported metrics keep their original conditions attached, so a number is never quoted without the caveats its author gave it.

  4. 04

    Records are revisited, not abandoned

    Each record carries its last-verified date and is re-checked on a rolling basis as repos move, demos break, or authors publish updated results. Corrections and takedown requests have a standing public policy, and the audit trail for every published record is kept.

The full evidence system — including where it falls short — is documented in How records earn their place.

Start with the strongest evidence

View all 151 →
E2 · artifact verifiedCommunitySearch Google Flights in seconds

A browser agent turns the visible DOM into an indexed action space and uses Jev to choose the next operation and compatible target.

E2 · artifact verifiedCommunityDrive macOS from structured screen state

A computer-use loop combines OCR and accessibility data, then asks Jev which bounded action should move the Mac toward a plain-English goal.

E2 · artifact verifiedCommunityChoose drone tactics from camera-derived state

A simulated quadrotor uses Jev for low-frequency tactical judgments while classical vision, flight control, and safety reflexes remain in code.

E2 · artifact verifiedCommunityComplete a StarCraft shareware mission

A reproducible harness lets Jev direct combat, exploration, and economy actions in the original StarCraft shareware campaign.

E2 · artifact verifiedCommunityTurn Home Assistant state into automation signals

A Home Assistant integration exposes Jev probabilities, choices, and scores as entities and action responses that automations can use.

E2 · artifact verifiedOfficialRoute smart-home assistant requests

TypeSafe's smart-home demo evaluates a request against many typed questions in parallel, then lets code use only the answers relevant to that request.

E2 · artifact verifiedOfficialBatch a regulatory briefing into one call

An official cookbook asks 13 regulatory questions over a pinned GDPR article in one Jev request and compares that batch with 13 separate requests.

E2 · artifact verifiedOfficialCheck LLM citations against source context

An official cookbook combines exact quote matching with a Jev relation judgment to label citations verified, unsupported, contradicted, or fabricated.

E2 · artifact verifiedCommunityPick Pokémon turns from live ROM state

A battle harness reads FireRed state from RAM and lets Jev choose the next legal move or switch while ordinary code advances the fight.

E2 · artifact verifiedCommunityAssess a simulated company under attack

A local cybersecurity lab replays synthetic telemetry and asks Jev for compromise probability, classification, severity, and an advisory response as evidence accumulates.

E2 · artifact verifiedCommunitySearch the web with typed intent judgments

A metasearch front end lets Jev choose the query, sources, and time range, then score every result for relevance while code fans out to engines.

E2 · artifact verifiedCommunityReplace Claude Code compaction with Jev decisions

A Claude Code plugin and npm library asks Jev which old tool calls and results still matter, then drops or truncates the stale ones, so /compact replaces the lossy built-in summary with the kept messages verbatim.

Curated paths through the ledger

Thirteen domains, one honest ledger

65 recordsDeveloper tools

Coding-agent supervision, review routing, and typed tool gates — the largest cluster in the index.

38 recordsAutomation

Browser, desktop, and smart-home control where Jev picks the action and deterministic code executes it.

36 recordsResearch and measurement

Measurement studies on labelled data, verbatim extraction, and metric conditions written out in full.

30 recordsAgent control

Referee patterns: gating risky tool calls, pruning agent history, exposing judgments over MCP.

17 recordsGames and interactive fiction

Emulator-state control loops that show typed decisions holding up at interactive frame rates.

14 recordsSecurity

Alert triage, CVSS mapping, and PII redaction built as narrow, checkable judgments.

14 recordsSearch and retrieval

Reranking and typed intent judgments over candidate sets that deterministic code assembles.

5 recordsSmart home

Home Assistant and domestic systems where one typed decision maps to one safe action.

5 recordsMobile

Android driven from the accessibility tree, Flutter bindings, and helpers that fail open.

4 recordsMusic

Note-by-note composition and improvisation under explicit human direction.

3 recordsTrading

The highest-stakes cluster in the index — read it with the strictest posture.

3 recordsRobotics

Drones, arms, and delivery robots: bounded choices meeting the least forgiving failure modes.

3 recordsLegal and civic data

Gazette screening and permit extraction with verbatim outputs code can re-check.

Start here

Two short reads before the index

ExplainerWhat is Jev?

The System One model that returns typed decisions instead of chat — explained with evidence, and with the other JEV disambiguated.

AnalysisJev vs. chat models

Output shape, responsibility boundaries, and failure modes — a comparison with no invented benchmarks, grounded in the indexed records.

From the desk

Original analysis and guides

MethodologyHow records earn their place

The five gates behind every record — public artifact, source classification, evidence levels, mandatory limitations, rolling re-checks — written by the editors who run them.

Ecosystem report · Oct 2026147 records, one pattern

Where the public Jev corpus clusters, which judgments people delegate, the architecture nearly everyone shares, and the holes nobody has filled — computed from the open dataset.

Tutorial · 15 minBuild your first decision loop

Observe state, ask Choice, Score, and Noul questions in one call, gate on confidence, act in code — the real API, the loop 147 projects share.

Open programReproduce a record, upgrade the index

Zero records have been independently reproduced. Run a public project, publish your conditions, and upgrade a record to E3 — contradictions count double.

Add to the record

Built something public with Jev?

Send the artifact, what Jev decided, and any measured outcome. Nothing appears in the directory before editorial review.

Submit evidence