Article
CommunityJev Guides & Articles
Vega: Jev gate closed up to a third of triage alerts
Vega's proof of concept put Jev in front of its alert triage agent on two tenants: it gated off 15% of alerts on the busy tenant and a third on the quiet one, about 98% correct by three LLM judges verified by its own analysts.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Author
- Ido Kuks
- Last verified
- Oct 2026
Used inJev for Security & Privacy
Highlights
- Vega put Jev in front of its alert triage agent on two tenants; as a gate it closed 15% of the agent's alerts on the busy tenant and a third on the quiet one.
- The gate was about 98% right by three LLM judges whose labels Vega's security analysts verified and corrected by hand, one cheap call each.
- Jev could not decide alerts outright, which was never the goal; the tenant's own history is what made the gate work, the post says.
- Vega's arithmetic puts Jev about 70 times cheaper per input token than a frontier model, with output free and questions run in parallel.
- Listed limits: Jev can't count, is bad at comparing numbers, negations confuse it, it reads literally, and state tops out around 32k tokens.
Watch out
Copy reviewed against the published post on 2026-10-17. The analysis is vendor-authored: Vega sells an agentic security platform, the proof of concept ran on two of its own tenants, and accuracy was judged by three LLM judges with its own analysts verifying the labels, so the figures are not independent.
Questions about this entry
Short answers drawn from the linked project or page.
Vega put Jev in front of its alert triage agent on two tenants as a gate: it closed 15% of the agent's alerts on the busy tenant and a third on the quiet one, right about 98% of the time, at one cheap call per decision.
Three LLM judges produced labels in one cheap call each, and Vega's security analysts then verified and corrected those labels by hand. The post's headline sums it up as 230x faster, 2,000x cheaper and 98% right.
No. The post says that was checked but never the goal: Jev cannot make the decision itself, and the tenant's own historical data about its alerts is what made the gate work in this proof of concept.
TypeSafe quotes 70 to 500 ms per call, and questions in one request are evaluated in parallel, so twenty questions take about as long as one. Vega's own arithmetic puts Jev about 70 times cheaper per input token than a frontier model, with output free.
It can't count, it's bad at comparing numbers, negations confuse it, it reads literally, and its state tops out around 32k tokens, so you can't feed it a novel. It fits steps where a model looks at data and picks from a list.
Vega is moving the same shape of decision across the platform: MITRE technique mapping, picking a query source, and the small decisions inside its agents for exploration, alert triage, incident investigation and detection engineering.
Spend a frontier model only where reasoning is needed and settle everything else with the smallest model that can, because, in Vega's framing, the outcome of an agentic SOC comes from how the models are put together around each decision.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
YouTube: Kubernetes admission control with a decision model, fail-closed.
Hacker News: SafeDep Gryph plus Jev checks for compromised coding-agent behavior.
2 pointsDetecting Compromised AI Coding Agents with Jev and GryphRead the thread on Hacker NewsX: Jev and Laya on security queries
X: Testing Jev for prompt injection
More Jev resources like this
Rogue Security's post uses Jev and System One models to frame agentic security as a decision problem: the same send_file action should be allowed, blocked or reviewed depending on who requested it and where the data goes, illustrated with policy case files rather than product results.
Jev Guides & ArticlesArticle#community#press#security
Community
A BIG CHANGE analysis of the Diogo Almeida Latent Space interview: Jev's most consequential proposition is the unit of automation, a bounded judgment inside an existing process, and the launch claim that Jev can't hallucinate needs a narrow reading because schema matching does not establish truth.
Jev Guides & ArticlesArticle#community#press#explainer
Community
VentureBeat's Louis Columbus reports Jev is spreading into agent infrastructure faster than enterprises can audit it. In one Octomind test a planted field cut a block probability from 0.76 to 0.48, and TypeSafe and Pydantic both warn inputs can steer answers.
Jev Guides & ArticlesArticle#community#press#security
Community
A community field guide to TypeSafe's Jev with one documented support-ticket call, runnable JavaScript and Python starters, a JSON directory, and links to independent evaluations.
Jev Guides & Articles#community#awesome-list#sdk
Community
AbdelStark/awesome-typesafe-jevHTML
AbdelStark/awesome-typesafe-jevHTMLA CLI that turns Jev into Unix pipeline primitives: is, choose, score, filter, and guard commands return typed verdicts, stable exit codes, and versioned JSON without prompt or parsing glue.
Jev Tools & CLIs#community#python#cli
Community
sharziki/semdecidePython
sharziki/semdecidePythonA one-hour, six-step workshop built around a fighting game that teaches TypeSafe's Discriminative (Jev) decision model and puts it next to Gemini in a Google ADK workflow. It runs in Cloud Shell or anywhere signed into gcloud, with hosted Jev, a local DiffusionGemma VM, or a rehearsal stand-in.
Jev Guides & Articles#community#github#workshop
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.