QUAICU
Fast, low-cost AI decisions
for your tickets and your AI stack.
We implement Jev, TypeSafe AI's classification model, inside your helpdesk and AI workflows: ticket classification and routing, choosing the right AI model for each request, guardrails and escalation. A person stays in the loop wherever the model isn't sure.
A model that decides,
not one that writes.
Jev reads text and answers structured questions about it in a single call. It returns probabilities rather than prose, so every decision can be thresholded, logged and checked.
- Yes or no
- The probability that something is true, for example “does this ticket need a person?”
- Choice
- One option from a list, with a probability for every option, for example which queue a ticket belongs to.
- Score
- A position on a scale of 2 to 10 levels, for example urgency or sentiment.
- Price
- $0.042 per million input tokens and no charge for output, on TypeSafe's published price list.
- What it doesn't do
- It doesn't write replies, code or summaries. Where text is needed, we pair it with a language model and let Jev decide when that model is worth calling.
Six places Jev
earns its keep.
Support ticket classification
Label every incoming ticket by type, product, urgency and language, and route it to the right queue. Tickets below a confidence threshold go to a person instead of being guessed.
AI cost control
Put Jev in front of your large language models. It decides which requests a small, cheap model can handle and which need a stronger one, or checks an answer before you pay for a second call.
Escalation to a person
Decide when a customer conversation should leave the bot and reach a human agent, based on intent, sentiment and risk.
Guardrails and moderation
Screen user input and AI output against your policies: allow, warn, review or block, with the probability recorded for every decision.
DPDP and privacy data mapping
Flag records, documents and messages that contain personal data, as a first step toward knowing where personal data lives before India's DPDP duties start in May 2027.
Inbox and alert triage
Sort shared inboxes and operations alerts: separate real support requests from noise, group duplicates and tag what needs action.
Enterprise rollout,
step by step.
- Agree the data route first
- Direct API or through Cloudflare, data retention, and what text is sent. Personal data is minimized before it leaves your systems.
- Agree the labels
- We design the categories with your team, with clear definitions and examples, and decide what a wrong answer costs for each one.
- Build a test set
- We label a sample of your real tickets or messages as the reference to measure against.
- Measure before go-live
- Jev runs in shadow mode next to your current process. We report precision and recall for every category, not one average.
- Set confidence thresholds
- High-confidence results are applied automatically; the rest go to a person. You choose the threshold per category.
- Connect your tools
- Webhooks from your helpdesk or app send the text; the result comes back as tags, fields or a routing decision.
- Log and monitor
- Every decision is recorded with its probabilities. We watch accuracy over time and update labels when your products or customers change.
Our full guide: Jev for enterprise: best uses and how to implement.
What others have measured.
Jack & Jill · candidate screening with Jev
Published by TypeSafe AI. Replacing part of an LLM screening step with Jev cut cost by 88% and median screening time per candidate from 20.3 to 10.3 seconds. The quality change was not statistically significant.
Microsoft Azure · DeepTriage incident routing
Peer-reviewed. An automated classifier that routes incidents to the right team reached 82.9% F1 and has been used in production by thousands of teams since 2017.
Accenture · automated incident assignment
Published by Accenture. Automated classification in ServiceNow assigns about 2,000 of roughly 10,000 daily internal IT incidents without a person.
These are published results from the companies named, linked to their sources. They are not QUAICU clients or QUAICU results.
Straight answers.
- What is Jev?
- Jev is a classification model from TypeSafe AI. Instead of writing text, it answers structured questions about the text you send: yes or no with a probability, a choice from a list of options with a probability for each, or a score on a scale. One call can answer many questions at once.
- What is Jev good for?
- Decisions about text: which team a ticket belongs to, whether a message needs a person, which AI model should handle a request, whether content breaks a policy, or whether a field an AI extracted is likely wrong.
- What can't Jev do?
- It does not write replies, code or summaries. TypeSafe's own documentation says it is weak at counting and exact numbers, that accuracy drops when unrelated material is included, and that score levels should not be read as exact distances. We design around these limits.
- How does Jev reduce AI costs?
- It is priced per input token with no charge for output ($0.042 per million input tokens on TypeSafe's published price list). Used as a first step, it can send simple requests to a cheaper model and only call an expensive model when needed, or check an AI answer before you pay for a second pass.
- How accurate is it on our tickets?
- That depends on your categories and data, so we measure it before go-live: we label a sample of your past tickets, run Jev alongside your current process in shadow mode, and report accuracy per category. Low-confidence results always go to a person.
- Does TypeSafe AI train on our data?
- TypeSafe's privacy policy says it does not train or fine-tune models on customer inputs. Zero data retention is offered to enterprise customers, and Jev is also available through Cloudflare, which lists it with zero data retention. We agree the route with you before any data is sent.
- Which tools can it connect to?
- Any system that can send a webhook or call an API: helpdesks such as Zendesk, Freshdesk, Jira Service Management and ServiceNow, your own apps, and AI gateways such as Cloudflare AI Gateway, Vercel and LiteLLM.
- Is QUAICU affiliated with TypeSafe AI?
- No. Jev is a product of TypeSafe AI. We use it in client work and implement it for teams that want it.
Jev is a product of TypeSafe AI. QUAICU is not affiliated with TypeSafe AI. Product details are taken from TypeSafe AI's public documentation and were last reviewed in October 2026.