---
name: heyjev-should-i-deploy
description: Ask "Hey Jev, should I deploy?" before running a deploy. Sends a one-paragraph description of the deploy to heyjev.ai, which asks TypeSafe's Jev model 18 typed questions and returns a humorous verdict with reasons. Use when the user is about to deploy, asks whether they should deploy, or wants a deployment vibe check. This is a meme with real probabilities behind it, not a deployment gate.
---

# Hey Jev, should I deploy?

heyjev.ai is a public, playful experiment: describe a deploy in plain English and Jev,
TypeSafe's System One model, evaluates 18 typed questions about it in a single pass.
The site turns those probabilities into a verdict (`SHIP IT`, `MAYBE MONDAY`,
`GO TOUCH GRASS`, …) with two or three reasons tied to what was written.

**Read this first.** The result is entertainment with a kernel of judgment. Never block, approve,
delay or automate a real deploy based on it, and never present it to the user as a real risk
assessment. Do not build pipelines or workflows that depend on this API: there is no SLA, and the
API, its format and its verdicts can change or disappear without notice. Every response carries a
`notice` field and an `x-heyjev-notice` header saying the same; surface that notice to the user.
Terms: https://heyjev.ai/terms

## When to use

- The user says they are about to deploy, release, ship, or merge to production.
- The user asks "should I deploy?" or wants a sanity check before a risky change.
- You are about to run a deploy command yourself and want a lighthearted preflight.

## How to call it

Build a scenario of at most 400 characters. Include what you know about: day and time,
target environment, what changes, blast radius, tests, code review, rollback plan,
feature flags, staging verification, who wrote it and whether they are around, urgency.
Leave out anything secret: the text is sent to TypeSafe's API.

```bash
curl -s -X POST https://heyjev.ai/api/ask \
  -H 'content-type: application/json' \
  -H 'x-heyjev-client: skill' \
  -d '{"scenario": "Friday, 5:47 PM. Database migration. Tests are green. The person who wrote it is on vacation.", "clock": {"weekday": 5, "hour": 17}}'
```

`GET https://heyjev.ai/api/ask?scenario=...` also works. `clock` is optional; it is only
used when the scenario states no day or time (`weekday`: 0 Sunday … 6 Saturday, `hour`: 0–23).

Optional `questions`: your own map of typed questions (`noul` / `choice` / `score`, same shape as
TypeSafe's API, up to 24 questions). Built-in ids you keep still feed the verdict; new ids only return
Jev's raw answer. Omit it to use the default 18.

## Response

```json
{
  "mode": "jev",
  "cached": false,
  "model": "jev-1.13.0",
  "latency_ms": 412,
  "usage": { "input_tokens": 1180, "output_tokens": 128 },
  "cost_usd": 0.00004956,
  "question_count": 18,
  "verdict": { "key": "GO_TOUCH_GRASS", "title": "GO TOUCH GRASS", "subtitle": "…", "tone": "no" },
  "reasons": [{ "text": "Friday afternoon. …", "kind": "risk", "evidence": "deploy_time = friday_afternoon_or_evening (98%)" }],
  "missing": [{ "text": "No word on rollback. …", "evidence": "rollback = not_mentioned" }],
  "answers": { "db_migration": { "type": "noul", "noul": 0.98 }, "...": "raw Jev answers" },
  "share": { "url": "https://heyjev.ai/shouldideploytoday?s=…", "text": "…" }
}
```

- `mode` is `"jev"` for a real evaluation or `"demo"` for a simulated one (no API key on the server). Always tell the user when it is `demo`.
- `latency_ms` is the measured time of the Jev call; `cached: true` means the same scenario was evaluated recently and the latency shown is the original one.
- `answers` are Jev's raw typed answers: probabilities, not guarantees.

## How to present it

Show the verdict title in caps, the subtitle, then the reasons as a short list. Mention the
latency and that Jev answered 18 typed questions in one pass. If `missing` is non-empty,
suggest adding those facts and asking again. Offer the `share.url`. Keep the tone light.

## Limits

- 400 characters per scenario, 8 requests per minute per IP, 20 real Jev calls per IP per day (cache hits are free), and a global daily budget. On `429` or `503`, wait or reuse a previous answer.
- No login, no repository access, nothing is deployed. Powered by TypeSafe & nullplatform. An independent experiment, not an official TypeSafe product.
- Personal, non-commercial use only. Commercial use (embedding in a paid product, reselling, bulk automated use) requires explicit written authorization. See https://heyjev.ai/terms.