Field notes / Jev
Small decisions.
Real possibilities.
Explore how Jev turns text into structured decisions, with practical examples, code, and community projects.
The input below is a sample. Open the message routing recipe to run this question yourself.
Experiment 01
Where should this message go?
No recorded answer for this sample yet. The message routing recipe runs this question for real, and the guide below shows the request and the shape of the answer.
Three ways to put Jev to work.
Route a message, score its urgency, or check whether a reply covers the request. Explore an example, then try your own.
- 01 / CHOICE
Route a message
Choose which department should handle a customer message.
Try message routingCustomer message
“Hi, I was charged twice for my subscription this month. The two charges show up on my card statement on the 3rd and the 4th. Can you refund one of them?”
- billing(charges)
- account(login)
- technical(bugs)
- sales(pricing questions)
- other(anything that does not fit the departments above)
Illustrative exampleDiagram only. Probabilities appear once a live or recorded result is shown.
- 02 / SCORE
Score urgency
Evaluate how urgently a message needs attention.
Try urgency scoringMessage
“Our largest customer's dashboard has been showing stale data since this morning, and they are presenting it to their board right now. Could someone look at this immediately?”
RoutineTime-sensitiveUrgentScore 1.82 of 2: a read against the rubric above, not a measurement.
Illustrative example - 03 / NOUL
Check a reply
Check whether a draft reply addresses the customer's explicit requests.
Try reply checkingCustomer
“Can you refund my last payment and send me a confirmation email once it's done?”
Draft reply
“I've processed a refund for your last payment. You'll receive a confirmation email at the address on file within the next few minutes.”
Addresses all explicit requestsDiagram only. A real answer returns one probability from 0 to 1.
A coverage check, not a fact check.
Illustrative example
Start building / Guide 01
Your first Jev request.
Route a support message with TypeScript. Define the departments, send the message, and use the returned decision.
TypeScript · Beginner
- 1
Provide the message
The text Jev evaluates.
- 2
Define the departments
The options it can choose from.
- 3
Use the decision
Handle the result in your application.
More guides
- How to try Jev: API key and first request
What works without a key, where to get one, and a first cURL request.
- Jev Python tutorial
Classify, score, and check a message with the typesafe-sdk package.
- Jev pricing and subscription questions
What a request costs, and what TypeSafe does and does not publish about billing.
const message = "Hi, I was charged twice for my subscription this month. The two charges show up on my card statement on the 3rd and the 4th. Can you refund one of them?";
const { answers } = await client.systemOne({
state: message,
model: "jev-latest",
questions: {
department: choice(
"Which department should handle this customer message?",
{
billing: "Charges, invoices, refunds, payment methods, subscription costs",
account: "Login, password, profile, permissions, closing or changing an account",
technical: "Bugs, errors, outages, integrations, something not working",
sales: "Pricing questions, plan comparisons, upgrades, purchasing for a team",
other: "Anything that does not fit the departments above",
},
),
},
});
const { choice: result, probabilities, confidence } = answers.department;{
choice: "billing" | "account" | "technical" | "sales" | "other",
probabilities: { … },
confidence: 0–1
}No recorded response yet. This is the shape only, not a result. See the guide for a full response.
Community
Built with Jev.
Three projects to start with. Every entry is checked against its own repository. Listing is not an endorsement.
Explore all projects- Coding agents
Jev Review
by devagrawal09
Runs a staged pipeline of Jev judgments over a git diff or a whole codebase and reports findings in a local dashboard, keeping every threshold in code.
View repository (opens in a new tab) - Automation and integrations
HA-Jev
by AboveColin
Turns Jev questions about Home Assistant entities into sensors and automation actions, with a daily token budget and per-day spend reporting.
View repository (opens in a new tab) - SDKs and developer tools
pg-jev
by realZachi
A PostgreSQL extension that lets plain-language conditions filter, rank, and classify table rows by sending batches to Jev, with no embeddings or vector column.
View repository (opens in a new tab)
Official resources
- TypeSafe documentationConcepts, the three question types, confidence, and patterns.DocsTypeSafe AIRead Sept 2026Read Sept 2026 (opens in a new tab)
- HTTP API referenceRequest and response fields for /v1/systemone, plus error codes.ReferenceTypeSafe AIRead Sept 2026Read Sept 2026 (opens in a new tab)
- typesafe-sdk-jsThe TypeScript client used in Guide 01. MIT licensed.SDKtypesafe-aiType-checked v0.6.0Type-checked v0.6.0 (opens in a new tab)