Prototype · closed beta
unOpen.AI
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
TypeSafe AI · Jev
Jev AI.
Build tools
that decide.
Jev is TypeSafe AI’s System One model for fast, structured decisions: it answers focused questions about a request with bounded outputs rather than long generated text. For product teams, we design tools that use those decisions to select an action, route a request or flag uncertainty, while your own software keeps control of permissions and every final action.
TypeSafe early access · Research checked 18 September 2026
Introduction transcript. Qubi: Jev answers typed questions fast, so your software can decide what happens next. Model routing: each request is assessed first, then sent to a model that fits. Agent guardrails: quick checks before an agent reads, answers or acts. Real-time agents: several questions at once, and a safe move if time runs out. Benchmarks: speed, cost and agreement, read together and tested on your own work.
The model
A decision engine inside your product.
A support tool needs to choose a queue. An agent needs to select a permitted function. A model router needs to decide whether a request warrants expensive reasoning. These are bounded judgements that feed the next line of code.
Jev accepts state and typed questions, then returns structured answers. TypeSafe describes its training approach as Reinforcement Learning for Calibrated Decisions. It is built for this machine-facing role; text generation remains a separate model’s job. TypeSafe’s model overview.
Our integration work starts with the tool you want to build: its inputs, allowed actions, failure cases and measurable success criteria. A model is useful when the whole workflow improves.
Tool building blocks
CHOICE
Choose an action
Pick from a defined set of handlers or tools. Choice returns the selected option, a probability distribution and confidence. Include a route for requests that do not fit.
SCORE
Score what matters
Describe levels of urgency, relevance or task complexity. Score returns a value and distribution across those levels, with confidence for the next gate.
NOUL
Check a condition
Ask whether a specific condition holds. Noul returns a probability from zero to one; it has no separate confidence field. Your code decides what that means for an action.
These are TypeSafe’s three documented primitives. Independent questions can share one request; an answer needed to construct the next question requires another call.
A tool you could build
One request. Several useful decisions.
Consider an internal service-desk tool. Pass the request and relevant account facts once. Ask which team owns it, whether it describes an outage, and how urgent it appears. The application combines those answers with the actual service contract and permissions.
A clearly classified outage can open a permitted incident workflow. An ambiguous request goes to a person. The model does not invent an API endpoint, grant itself access or calculate the customer’s entitlement. Those rules stay in the application.
This is our proposed integration pattern, informed by TypeSafe’s function-calling cookbook and parallel-question pattern. We would test it on your historical requests before enabling automatic actions.
From judgement to action
PROPOSED TOOL ARCHITECTURE
-
01Application state
request + relevant facts
-
02Jev questions
intent · urgency · fit
-
03Decision policy
thresholds + permissions
-
04Approved tool
validated arguments
-
05Record outcome
audit + evaluation
What we can build with it
A model router that respects your budget
We are building a router that assesses the work before selecting a model. Reserve expensive reasoning for requests that need it, and measure the quality and cost of every route.
Explore model routing
Checks around an autonomous agent
Screen retrieved content, assess proposed actions and escalate uncertain cases. Keep access controls and approval requirements in deterministic code.
Explore agent guardrails
Responsive agents and game loops
React to structured state with several decisions at once. Learn what TypeSafe’s game demo establishes and how to budget the full response loop.
Explore real-time decisions
Evidence for the model decision
Compare published latency, cost and agreement. Understand the evaluation conditions before translating a provider’s chart into a product requirement.
Read the benchmark comparison
Published specification
- Provider / model
- TypeSafe AI · Jev 1.13 · versioned ID jev-1.13.0
- Input
- Text: strings, JSON objects or arrays of text values. No direct image, audio or video input.
- Token budgets
- 64k total tokens per request; state plus the longest question must fit within 32k.
- Published price
- US$0.042 per million input tokens. Output tokens are free at the time of writing.
- Access
- Hosted API; early access. Confirm account access and current limits with TypeSafe.
Provider-published specifications as of 18 September 2026, not figures we measured. Model families move quickly — check the provider's own documentation before committing to a number.
Source: TypeSafe model reference. Availability: launch announcement. This page offers QQuantum.ai integration work, not a TypeSafe account or an affiliation claim.
Where it fits
Fast decisions still need good engineering.
Use Jev where the answer space is bounded and the surrounding workflow can check the result. Keep arithmetic and date comparisons in code. Use a generative model for writing, code generation and open-ended synthesis.
Type-safe output can still be the wrong decision. TypeSafe documents weaknesses involving numeric precision, indirect questions, irrelevant context and adversarial content. A pilot needs representative examples, explicit fallback behaviour and evaluation after model changes. Published limitations.
Questions
Questions about Jev: AI Tools & Fast Decisions.
What is Jev AI?
Jev is TypeSafe AI’s System One model. It evaluates typed questions against supplied state and returns structured decisions that software can use. It is not a general text-generation model.
Can Jev make multiple decisions at once?
Yes. Independent questions can be evaluated in parallel against the same state. Their answers do not become context for each other. A dependent step may need a separate request.
Can QQuantum.ai help us build a Jev tool?
Yes. We can scope the decision, design the allowed actions and fallback path, and build an evaluation-led integration. TypeSafe account access and suitability for your data must be confirmed during scoping.
