Jev — Lab Component
Lab component — not in the Synedrix OS tech stack·Evaluated under LAB-003·Not required in any deployment mode
Not a Synedrix OS platform. Jev is not one of the twelve platforms and holds no OS layer. It is evaluated under
LAB-003alongside Laya, in structural isolation from production, as a candidate System One decision model. This page is lab reference material, not a stack platform page, and not citable as authority by any canonical document.
What Jev is
Jev, from TypeSafe AI, is a System One decision model offered as a managed API. You send it a piece of state
and one or more typed questions, and it returns one answer per question with a probability. It generates no text.
The model itself is proprietary and runs on TypeSafe’s infrastructure; the Python client, typesafe-sdk, is
published under the MIT licence.
| Concept | What it is |
|---|---|
system_one | The one call: state plus named questions in, typed answers out |
| Question types | Choice (a label and a probability for every label), Score (an expected score over an ordered rubric), Noul (the probability of yes) and Box (a location in an image) |
| Confidence | Every Choice and Score answer carries a confidence from 0 to 1, for code to compare with a threshold |
| Client | typesafe-sdk for Python, with retries and timeouts built in; the base URL is configurable |
What Jev is normally used for
- Routing and triage — choosing a queue, agent or model for a request
- Gating and verification — yes or no checks on inputs, tool calls and generated outputs
- Scoring — rubric scores where a calibrated probability matters more than an explanation
The common thread: replacing an LLM call at a bounded decision point with a cheaper, faster call that returns a probability instead of a sentence.
How Synedrix OS uses it
Synedrix OS does not use Jev in production. Under LAB-003 it is called from the lab’s isolated environment
only, and only synthetic or already-public data is ever sent to it, because a managed API means the data
leaves the network.
What it owns
Nothing in production. Inside the lab, it answers the same questions as Laya on the same labelled decisions, as the reference point for Laya’s quality.
What it does not own
| Concern | Owner |
|---|---|
| Whether any action is permitted | Policy Guard Agent (os-agt-002) and HITL Gateway Agent (os-agt-004). A decision model’s confidence is evidence, never permission |
| Committing durable memory | Memory Curator Agent (os-agt-010) |
| Model routing policy for registered agents | The agent’s runtime policy (Agent Runtime Policy Spec) |
No registered agent may hold Jev, or a tool wrapping it, in its tools_allowed.
Why Jev is evaluated, and the limits on adopting it
- It is the managed reference point. Jev needs no local hardware, which makes it the quickest way to measure what a System One decision model can do for Synedrix-shaped decisions.
- It cannot serve an air-gapped deployment. Mode 3 forbids any outbound model call, so Jev can never be the default decision model for Synedrix OS.
- Its data terms are part of the evaluation. TypeSafe’s data-processing, retention and pricing terms are not yet verified and are read and recorded under
LAB-003.
Official documentation
- TypeSafe AI
- TypeSafe documentation
typesafe-sdkon PyPI — Python client (MIT)
Synedrix OS · Lab component reference · Jev