articlesource-backed
Building a harness with Jev
Architecture notes on putting bounded Jev judgments inside an agent harness.
ROUTING, GATING, VERIFYING
Agent harnesses
Where a typed decision sits inside a larger loop: model routing, tool-risk gates, context compaction, rollout rules.
The same pattern across sources: a generative model writes, Jev decides, code executes.
66 entries
articlesource-backed
Harness design with Jev
A long-form explanation of integrating typed judgments into agent workflows.
postcommentary
A constrained primitive for harnesses
Jev is framed as a constrained structured-output primitive inside a larger agent harness.
demoauthor demo
Agent routing in FirstMate
A model or agent route is chosen from a fixed catalog based on the current task.
reposource-backed
Jev Codex Router
Per-turn routing chooses a model, reasoning depth and speed mode for Codex.
demoauthor demo
Instant context compaction
Jev scores which tool calls and results remain relevant before context is compacted.
postcommentary
Four decision-layer directions
A conceptual map of where a low-latency decision model can sit in software.
postcommentary
Four rules for production rollout
Keep a bypass, decompose questions, tie confidence to actions and shadow-run before replacement.
articlecommentary
Jev engineering in ten steps
A community guide covers typed calls, batching, middleware, confidence gates and migration.
postcommentary
Find every decision in a workflow
A prompt asks coding agents to inventory every bounded semantic decision where Jev might replace a heavier model call.
reposource-backed
Building with Jev skill
An agent skill for finding and designing bounded Jev decisions in software.
integrationsource-backed
Jev on Vercel AI Gateway
Vercel's changelog entry: Jev is callable through AI Gateway from AI SDK 7 (7.0.105+) via the experimental `evaluate` API as model `typesafe-ai/jev`, with Choice, Score and Noul questions.
integrationsource-backed
Jev 1.13 on OpenRouter (beta)
OpenRouter's beta listing for `typesafe/jev-1.13` with a `jev-latest` alias: $0.042/M input, free output, 32K context, plus the platform's own latency and token-volume telemetry.
integrationsource-backed
Jev on Cloudflare Workers AI
Workers AI exposes the model as `typesafe/jev` (jev-1.13.0) with `state` plus noul / choice / score questions and a 32,000-token context window.
integrationsource-backed
Jev in Netlify AI Gateway
Netlify's changelog: the gateway injects credentials so `@typesafe-ai/sdk` runs inside Netlify Functions without managing keys; alias `jev-latest`, Node 20+.
demoauthor demo
Vercel's fx safety classifier on Jev
Vercel benchmarked the auto-mode safety reviewer of its fx coding agent with Jev in place of GPT Luna and says it is likely to become the default.
reposource-backed
toolgate: a tool-call firewall for agents
Screens agent tool calls for destructive, exfiltration, privilege-escalation and off-task actions using Jev probabilities, ships as an npm package and a Claude Code PreToolUse hook, and logs an audit trail.
reposource-backed
jev-mcp
A proof-of-concept MCP server exposing three tools: verify a claim against evidence, screen text for injection or relevance, and rank candidates semantically.
reposource-backed
jev-rules for Claude Code
A Claude Code plugin that asks Jev which standing rules and codebase maps are relevant to the current prompt or file edit, injects only those, and falls back to all rules if the API is down.
demoauthor demo
Claude Code hook routing tasks with Jev
A Claude Code hook where Jev decides whether a task goes to Haiku, an Opus sub-agent, or an external harness such as Codex or OpenCode.
articleauthor demo
Empryo: Jev inside a coding-agent harness
Wires Jev into five decision points of a coding-agent harness; the main test is failure triage on 102 real provider-API failures from session logs.
articlesource-backed
Langfuse: using Jev for evals
A tutorial that replaces an LLM-as-judge with Jev to score Langfuse traces, with a full Python example detecting user disagreement per conversation turn.
benchmarksource-backed
Jev as an LLM router: a mostly negative result
Benchmarks Jev as a router across 13 flagship models on LLMRouterBench and a RouterArena pilot.
articleauthor demo
Replacing model-routing classifiers with Jev
Uses Jev as a four-tier difficulty classifier in place of NVIDIA NeMo Switchyard's router classifiers, with code.
integrationsource-backed
hono-jev-router
An experimental semantic router for Hono by its creator: routes are described in plain language and Jev matches incoming requests to them from method, URL, headers and body.
integrationsource-backed
n8n community node for the System One API
An n8n node with Ask Yes/No, Ask Choice, Ask Score, batch Evaluate and model lookup operations, defaulting to `jev-latest`.
reposource-backed
jev-shield: semantic MCP firewall
Sits between MCP clients and servers with three layers: structural allowlists, semantic hazard detection via Jev, and session taint tracking; targets poisoned tool descriptions, injection, exfiltration and secrets.
integrationsource-backed
tiershift: policy-bounded LLM cost router
Jev scores each request on eleven dimensions (difficulty, stakes, safety and more) and YAML thresholds map the scores to model tiers; TypeScript and Python SDKs plus an HTTP proxy.
integrationsource-backed
jev-router on LiteLLM
An OpenAI-compatible router that uses Jev to pick a model on top of LiteLLM, filtering candidates by capability and falling back to the cheapest model without a TypeSafe key.
reposource-backed
jev-router for Claude Code
Routes Claude Code tasks to a cheaper or more suitable model based on a Jev decision; listed in the Studio Yebisu roundup.
integrationsource-backed
Jev as a prompt guard in agentgateway
The agentgateway proxy project ran Jev as a prompt guardrail with distributed tracing and cost tracking, with an example linked.
postauthor demo
Jev as a pre-action safety monitor
Tested Jev as a pre-action safety check for agents; the follow-up setup filters everything with Jev and escalates only suspicious cases to Gemini.
postsource-backed
Foreman: a supervisor for coding agents
Foreman watches coding agents and uses Jev to continuously assess progress, completeness, tests, drift and verification, intervening when needed.
integrationauthor demo
skillbox: skill routing with Jev
skillbox reads the query and returns the relevant agent skills immediately with Jev, instead of agents spending many turns searching for them.
integrationsource-backed
ruby_llm-typesafe
Adds a `:typesafe` provider to RubyLLM 2 with Noul, Choice and Score primitives, schema building and validation, and an offline models catalog.
integrationsource-backed
typesafe-ai: Rust bindings for the System One API
A Rust crate with enum responses and confidence, async or blocking backends, batched questions and observable retries; examples cover PR questions and semantic search.
integrationsource-backed
advocaat: a small type-safe client
A TypeScript client for batching several typed questions about your data into one Jev request; the example triages GitHub issues.
integrationsource-backed
jod: semantic schemas over Jev
Validates state locally with Zod-style schemas, bundles all questions into one parallel Jev request and projects typed answers; built on the official SDK.
demosource-backed
LangGraph email workflow with Jev
An async LangGraph workflow that classifies mock emails (invoice versus general) with the Choice primitive and routes them to handlers, with unit tests; pins jev-1.12.
integrationsource-backed
Pydantic AI: first-class TypeSafeModel
Pydantic AI merged a TypeSafe model provider: `TypeSafeModel('jev-latest')` with confidence thresholds, a FileUnderstanding capability and fallback to an LLM for tool calls; docs cover jev-latest, jev-preview and pinned releases.
integrationsource-backed
LiteLLM: Jev-driven context compaction and pass-through
The LiteLLM proxy scores older tool results with Jev and replaces low-relevance ones with short removal notices before the request reaches the LLM; a TypeSafe pass-through endpoint is documented too.
integrationsource-backed
Braintrust: Jev as a typed scorer for agent responses
Braintrust added Jev as a scorer with a confidence-threshold pattern (accept above 0.95, route 0.70–0.95 to an LLM judge, flag below 0.70 for humans) and auto-instrumentation of TypeSafe client calls in its Python SDK.
integrationsource-backed
OpenInference instrumentation for TypeSafe (Arize)
A Python package that traces TypeSafe System One calls as OpenInference LLM spans (request, response, model, token counts) for Phoenix and compatible backends; a JS package is documented in Phoenix.
reposource-backed
jev-pruner: trim tool output before the LLM sees it
A Claude Code and Codex CLI plugin that intercepts Bash output above roughly 10K estimated tokens, chunks it and asks Jev which lines are still needed, dropping progress logs and boilerplate.
integrationauthor demo
invalidate: memory-lease invalidation with Jev
A layer that decides when an agent's stored memory should be invalidated, using Jev checks; a playground is hosted.
integrationauthor demo
Agent Handoff Gate
An experimental protocol where a worker agent sends requirements, code and test results to Jev, which returns an evidence verdict for the team-lead agent before results are accepted.
integrationauthor demo
mcp_jev: local MCP server with Jev packs
A local MCP server that runs Jev "packs" (code audit, PR review, GUI automation) for Cursor, Claude Code and Codex.
demoauthor demo
Flue agent + Jev via Cloudflare AI Gateway
A Cloudflare Worker demo where a Flue agent uses Jev through Cloudflare AI Gateway for intent and urgency routing.
integrationsource-backed
ruby_decision_model gem
A zero-dependency Ruby client for decision models supporting TypeSafe's native API and OpenRouter, with retries, timeouts and usage/cost fields.
integrationsource-backed
Jev as a decision layer for Grok Bot
A seven-step setup that puts Jev between Grok Bot and its actions: a jev-usage-router skill is consulted before browser, research, retry or extra-bot steps; it starts in shadow mode with logs, then goes active, with a kill switch. Repository included.
postcommentary
Jev as a harness primitive: evals, verifiers, self-improvement
Notes from integrating Jev into a custom agent harness: beyond cheap classification, the interesting uses are continuous evals and monitoring, custom verifiers for long-horizon agents, and recursive self-improving harness loops; a guide is in progress.
integrationsource-backed
hermes-jev-plugin: decision tools for Hermes Agent
A small plugin exposing Jev to Nous Research's Hermes Agent as tools: jev_check, jev_route, jev_score and jev_evaluate.
integrationsource-backed
typesafe-mcp: Jev as an MCP server for a local GLM agent
An MCP server for Jev built in about four hours and paired with a local GLM 5.3 Flash model — the author's 'absolute deadly combo'.
integrationsource-backed
quackd: one CLI for all your robots, Jev for the cheap steps
A multi-robot harness CLI (Microduck, Open Duck Mini, LeRobot, XLeRobot, AlohaMini and more) where each robot has an LLM brain and Jev handles the cheaper decision steps; a LeRobot SO-101 setup takes about five minutes.
integrationsource-backed
semgate: semantic request filtering for Go HTTP
Go HTTP middleware that lets Jev judge and route requests, plus a CTF-style example site inviting people to try injection attacks against it.
integrationsource-backed
Zammad ticket dispatcher on Jev
New Zammad ticket comes in, Jev decides which team owns it with a confidence score attached, and routes it.
demosource-backed
triage-bot: real-time support triage with Jev and Cerebras
Jev routes each ticket to a specialist agent (general / account / billing / technical) and decides whether a human should take it instead — all typed — while Cerebras writes the reply.
integrationsource-backed
jevlogs: score log signal before the expensive LLM pass
Open-source log triage for OpenTelemetry: Jev scores each log line's signal so only what matters reaches the expensive LLM analysis; a write-up covers Jev vs a fine-tuned alternative.
officialofficial
LangChain webinar: building a harness with Jev
LangChain's webinar on using Jev to build a better agent harness, with Sydney Runkle, Hunt Lovell and Allie.
articleauthor demo
Three days of Jev on production systems: map, shadow, measure, promote
No demo: notes from three days running Jev inside production systems, misses included, organised as map → shadow → measure → promote.
integrationsource-backed
hermes-jev-skills: routing, memory, compaction, GUI clicks for Hermes
Jev-powered model routing, memory, context compaction, skill selection and computer/browser use for Hermes agents (also Claude Code and Codex), with a routing dashboard and shadow mode that shows what Jev routing would have saved before you switch it on.
integrationsource-backed
jev-sift: classify first, read selectively
A portable agent plugin and MCP tool for batch text classification: the agent asks Jev whether a file is worth reading, a tool worth calling, a website worth browsing — before spending tokens on it.
integrationsource-backed
typesafe-mcp: Jev in Claude Code, Claude Desktop, Codex and Pi
An MCP connector that gives any MCP-capable agent direct Choice / Score / Noul calls to Jev; the Chinese roundups call it the first thing to install after getting an API key.
integrationsource-backed
SemDecide: typed semantic decisions for Unix pipelines and CI
Jev as a command-line filter: classify, score and filter lines in a shell pipeline, for crawlers, CI gates and data pipelines.
reposource-backed
Reticle: machine-native runtime perception for web and desktop apps
Agents can write code but struggle to understand what they built at runtime; Reticle gives web and desktop apps a Jev-style runtime perception layer so agents know which interface they are looking at.
postcommentary
Jev-as-a-judge for RL environment verifiers
Many RL tasks need a judge to verify pieces of outputs or trajectories, and at scale that is the bottleneck; a fast, cheap, calibrated judge could remove it. Draws on earlier work with Harvey's LAB benchmark, where harness engineering plus open models cut judging costs by orders of magnitude.