应用产品
Jev Playground 是一个用于测试和体验 Jev 的在线环境。
Jev Playground
<span class="tag">{provider === 'vercel' ? 'experimental_evaluate' : provider === 'typesafe' ? '/v1/systemone' : 'alpha.decisions'}</span>链接指向我们抓取当天的那个 commit,行号是准的。
下面是从这个项目源码里原样取出来的 question 组合。
category选项Main category of this ticket
bug — Product failure or unable to usebilling — Charges, payments or refundsfeature — Feature requestspam — Spam or meaningless contentbug_severity打分If this is a bug, how severe is it
0 — cosmetic: minor visual issue1 — minor: has a workaround2 — major: core feature affected3 — blocking: completely unusablewants_refund是/否Is the user asking for a refund?
frustration打分The user's mood
0 — calm1 — annoyed2 — furiousimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
category: { type: 'choice', instructions: 'Main category of this ticket',
criteria: { bug: 'Product failure or unable to use', billing: 'Charges, payments or refunds', feature: 'Feature request', spam: 'Spam or meaningless content' } },
bug_severity: { type: 'score', instructions: 'If this is a bug, how severe is it',
criteria: ['cosmetic: minor visual issue', 'minor: has a workaround', 'major: core feature affected', 'blocking: completely unusable'] },
wants_refund: { type: 'boolean', instructions: 'Is the user asking for a refund?' },
frustration: { type: 'score', instructions: 'The user\'s mood',
criteria: ['calm', 'annoyed', 'furious'] },
},
});fine选项Most specific issue type
duplicate_charge — Charged more than once for one purchasedeclined_card — Payment declined at checkoutrefund_status — Asking where a refund isinvoice_request — Wants an invoice or receiptcoarse选项Broad department
payments — Anything about charges, cards or refundsaccount — Login, profile, settingsproduct — Using the product itselfimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
fine: { type: 'choice', instructions: 'Most specific issue type',
criteria: { duplicate_charge: 'Charged more than once for one purchase', declined_card: 'Payment declined at checkout', refund_status: 'Asking where a refund is', invoice_request: 'Wants an invoice or receipt' } },
coarse: { type: 'choice', instructions: 'Broad department',
criteria: { payments: 'Anything about charges, cards or refunds', account: 'Login, profile, settings', product: 'Using the product itself' } },
},
});fit打分How well does this company match a mid-market logistics customer
0 — poor1 — partial2 — good3 — idealurgency打分How soon do they intend to buy
0 — someday1 — this year2 — this quarter3 — this monthhas_budget是/否Do they state that budget is approved or available?
decision_maker是/否Does the writer appear to be a decision maker rather than a researcher?
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
fit: { type: 'score', instructions: 'How well does this company match a mid-market logistics customer',
criteria: ['poor', 'partial', 'good', 'ideal'] },
urgency: { type: 'score', instructions: 'How soon do they intend to buy',
criteria: ['someday', 'this year', 'this quarter', 'this month'] },
has_budget: { type: 'boolean', instructions: 'Do they state that budget is approved or available?' },
decision_maker: { type: 'boolean', instructions: 'Does the writer appear to be a decision maker rather than a researcher?' },
},
});model选项Pick the most suitable model tier for the latest user turn
cheap — Everyday Q&A, typo fixes, light rewrites, small talkstrong — Multi-step reasoning, writing or debugging code, long-document analysis, mathnone — Greeting or acknowledgement that needs no model at allimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
model: { type: 'choice', instructions: 'Pick the most suitable model tier for the latest user turn',
criteria: { cheap: 'Everyday Q&A, typo fixes, light rewrites, small talk', strong: 'Multi-step reasoning, writing or debugging code, long-document analysis, math', none: 'Greeting or acknowledgement that needs no model at all' } },
},
});