Unofficial field guide to TypeSafe AI's Jev
What people build with Jev, and how to build it
Jev is a model that answers typed questions about your data (pick one, rate it, yes or no) with probabilities instead of text, in milliseconds. Below: 98 real projects, rated by Jev itself, and guides with working code.
How Jev fits into an app
Customer
You ask Jev
- Which queue?Choice
- How frustrated?Score
- Is it an outage?Noul
Jev
Jev answers
- Billing95% sure
- 0.45 of 2
- No2% chance
Billing team
your app routes it
not sure? a person decides
from typesafe_sdk import Choice, Noul, Score, TypeSafeClient
client = TypeSafeClient() # reads TYPESAFE_API_KEY
response = client.system_one(
state={
"subject": "Charged twice this month",
"body": "Hi, I see two charges of $49 on my card for September. Can you refund one?",
},
questions={
"queue": Choice(
instructions="Which support team should handle the ticket in `subject` and `body`?",
criteria={
"billing": "Charges, invoices, refunds, payment methods",
"technical": "Bugs, errors, outages, API problems",
"other": "Anything that does not fit the teams above",
},
),
"frustration": Score(
instructions="How frustrated is the customer in `body`?",
criteria=["Calm", "Frustrated", "Very angry"],
),
"is_outage_report": Noul(
instructions="Does `body` report that the product is down right now?",
),
},
)
queue = response.answers["queue"] # .choice == "billing", .confidence == 0.90
if queue.confidence >= 0.8: # your code decides, not the model
route_to(queue.choice) # your own function
else:
send_to_a_person(response) # your own functionimport { choice, noul, score, TypeSafeClient } from "@typesafe-ai/sdk";
const client = new TypeSafeClient(); // reads TYPESAFE_API_KEY
const response = await client.systemOne({
state: {
subject: "Charged twice this month",
body: "Hi, I see two charges of $49 on my card for September. Can you refund one?",
},
questions: {
queue: choice("Which support team should handle the ticket in `subject` and `body`?", {
billing: "Charges, invoices, refunds, payment methods",
technical: "Bugs, errors, outages, API problems",
other: "Anything that does not fit the teams above",
}),
frustration: score("How frustrated is the customer in `body`?", ["Calm", "Frustrated", "Very angry"]),
is_outage_report: noul("Does `body` report that the product is down right now?"),
},
});
const { queue } = response.answers; // queue.choice === "billing", queue.confidence === 0.9
if (queue.confidence >= 0.8) {
routeTo(queue.choice); // your code decides, not the model
} else {
sendToAPerson(response);
}curl https://api.typesafe.ai/v1/systemone \
-H "Authorization: Bearer $TYPESAFE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jev-latest",
"state": {
"subject": "Charged twice this month",
"body": "Hi, I see two charges of $49 on my card for September. Can you refund one?"
},
"questions": {
"queue": {
"type": "choice",
"instructions": "Which support team should handle the ticket in `subject` and `body`?",
"criteria": {
"billing": "Charges, invoices, refunds, payment methods",
"technical": "Bugs, errors, outages, API problems",
"other": "Anything that does not fit the teams above"
}
},
"frustration": {
"type": "score",
"instructions": "How frustrated is the customer in `body`?",
"criteria": ["Calm", "Frustrated", "Very angry"]
},
"is_outage_report": {
"type": "noul",
"instructions": "Does `body` report that the product is down right now?"
}
}
}'
# -> { "answers": { "queue": { "choice": "billing", "confidence": 0.90, ... },
# "frustration": { "score": 0.45, ... },
# "is_outage_report": { "noul": 0.02 } } }Use cases
Built by @twid · on X
This week my Discord bot started running every yes/no and "which one?" decision it makes through @typesafeai's Jev before anything else touches it.
Thoughts on Jev. Decisions like: Is this a request for the weather? Is this a request for image generation? Is this a request for…
AgentsSee the post ↗
Built by @ssbrouhard · on X
One System One call takes the ask + source and returns structured judgments:
Next layer on my @grok @bot build: typed triage with @typesafeai Jev sitting in front of the Grok Ship by @kunchenguid vs direct-handoff split. The problem I’m trying to solve is routing the wrong…
AgentsSee the post ↗

Built by @k0nOO · on X
The judgments go to plain code that picks the next question and which box to name.
My voice interviewer can't see the whiteboard; GPT-Live takes no images. The box I named "Fastly" is read as a CDN at 1.00, and my note "sync or async transcode?" as a question I still owe. Jev…
AgentsSee the post ↗
Guide with code · Advanced
Content Moderation and Injection Screening
Screen user messages and retrieved passages before they reach your LLM. One Jev request returns a policy Choice plus Noul checks for jailbreaks and injections.
Moderation and safetyRead the guide →
Built by @jack_gor · on X
I wired TypeSafe Jev into Codex and ZCode for model routing — not chat, just cheap decisions.
Most of what agents spend money on is not “writing code.” It is the selection loops around the work: which model to use this turn, whether a tool result matters, whether to open a…
AgentsSee the post ↗
Built by @zhongming_yu · on X
We tested Jev as a code reranker on 100 GitHub issues.
Model-planned grep + Jev was our best route, though the choice of retrieval route mattered more than the choice of reranker. The benchmark is CodeNib Base: 25 repos across C/C++, Go, Python, Rust and…
BenchmarksSee the post ↗

Built by @DokasIoann59897 · on X
I got early access to @typesafeai jev and built a demo with it based on a of…
RoboticsSee the post ↗
Guide with code · Intermediate
Email Triage and Priority Scoring
Sort a shared inbox with one Jev request. A Choice sets the category, three Score questions rate urgency, sender signals and effort, and code weights them.
Customer supportRead the guide →

Built by @dannylivshits · on X
What I built - a proof of concept for suspicious email classification aimed at scam emails: Scam Signal Lens.
I was exploring Jev by @typesafeai AI for safety use cases and it is an impressive and useful model, however don't expect it to magically auto classify any data without proper parameters and…
Moderation and safetySee the post ↗

Built by @LakshOnline · on X
So i wired typesafe's jev into a real pipeline and pointed it at the real api.
most jev demos i've seen are mockups with illustrative numbers. from an actual benchmark run: 7 of 9 items never reached an expensive stage. → one call per item returns category, relevance (0-2) and…
AgentsSee the post ↗

Built by @Chrisondesk · on X
I moderated 100 comments in real time for $0.002171.
A fifth of a cent. But this isn't a post about the demo. It's about the choice you're making when you add AI to a product: LLM or decision model. The model is Jev, by TypeSafe AI. Not a chatbot. A…
Moderation and safetySee the post ↗
Guide with code · Advanced
Hierarchical Product Categorization
Place marketplace listings in a taxonomy with thousands of leaves using cascading Jev Choice questions, a small beam, and confidence-based parent back-off.
Search and dataRead the guide →
Learn the basics
What is Jev?
System One vs LLMs, the three primitives, confidence, pricing, and limitations.
Jev vs LLMs for classification
Cost and latency math, and an honest list of when an LLM is the better tool.
Confidence thresholds
How to gate automated actions on confidence and tune the thresholds on your data.
Go to the source
This is an unofficial community site. API details change, so treat theofficial TypeSafe documentation as the source of truth, starting with theHTTP API reference and the list of known Jev limitations.