User repository on GitHub
A small, extensible decision-to-action harness for TypeSafe Jev
- Forks
- 2
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
Super Jev
An experimental TypeScript harness that connects evidence to Jev judgments, permitted actions, and verified outcomes, with pluggable data sources and tools, local JSONL traces, and zero runtime dependencies.
- Stars
- 14
- Language
- Python
- License
- MIT
- Reading time
- 4 min
- Published by
- Community
- Author
- Kevthetech143
- Last verified
- Sep 2026
About this project
One judge (Jev) between your agent and your data: the agent asks in its own words, Super Jev finds the file, checks the claim, permits the action, and remembers what you approved. Your agent stays responsible for the answer. One judge sits between your agent and your data: the agent asks in its own words, Jev finds the file, checks the claim, permits the action, and remembers what you approved. Your agent stays responsible for the answer.
Quoted from the project's README. Read the full README.
Highlights
- One bounded run tracks state, fetches evidence through observe, batches typed questions, validates answers, permits one tool, and verifies the outcome.
- Runs end as success, stopped, blocked, or error, bounded by maxSteps, timeoutMs, maxRequestBytes, and maxRepeats.
- The byte budget is a local size guard, not a tokenizer, and retries are not automatic; an action_started without action_completed needs reconciliation.
- Journals record context, answers, and outcomes; credentials stay out of trace payloads and raw error text is omitted.
- The suite has 50 tests, and a live jev-1.13.0 response is checked in as a fixture that confirmed the documented score rule and disproved one inferred rule.
Quickstart
ts
import { run, Jev, JsonlJournal } from './src/index.ts';
import { recovery } from './examples/domains.ts';
const result = await run({ domain: recovery(), initial: { restarted: false, healthy: false },
evaluator: new Jev(), journal: new JsonlJournal('runs/my-run.jsonl'),
maxSteps: 10, timeoutMs: 30_000, maxRequestBytes: 100_000, maxRepeats: 2 });
console.log(result.status, result.state);Watch out
MIT-licensed but experimental V0.2.0 and not affiliated with TypeSafe AI; needs Node.js 24+ and TYPESAFE_API_KEY, plugins run in-process as trusted code without sandboxing, and there are no automatic retries.
Questions about this entry
Short answers drawn from the linked project or page.
No, and that is the design rule: the cache fills only from your own approvals. Auto-save needs the same file to win the same question for you N times in a row on complete searches, plus a content check, and the secret scan and unchanged-file check still run; --approve meets the threshold at once. Every cache hit prints who approved it, --miss on a cached question un-saves it, and files the secret scan holds are never sent or saved.
No, and the README is explicit that version 1.0 does not promise semantic cache matching. Matching is the same question after lowercasing, collapsing spaces, and dropping trailing punctuation, with nothing fuzzier than that. A saved answer then lasts until its source file changes, with no clock expiry, and any listed file that has changed is withheld as STALE and searched live instead.
No, and that is listed under what 1.0 does not do: an agent stays in the seat for the final call, with you running the ask, read, answer, miss, or add loop yourself. It also does not sync automatically (data gets in through connectors you run; nothing watches your folders) and does not browse the web. Platform support is macOS and Linux, with Windows untested.
The maturity table is unusually blunt. ask, skill search, file navigate, connect and bulk prepare, the check gate, the permit gate, and the harness loop are LIVE, meaning proven in daily use. sweep, catalog, and fetch are MEASURED with live bench scripts, and fetch's admission gate is explicitly not yet met. chain-cli, derive-facts, investigate, passage-level search, saved-answer approve reuse, verify, and auto-catch are EXPERIMENTAL: moved aside, not deleted, not shipped.
Sent to the TypeSafe provider: file text when ask reads a file to confirm an answer, descriptions and questions used to rank files, and the claim plus evidence you pass to check. Files held by the secret scan stay local, and --writer builtin sends nothing while connecting unless you add --findability.
A state directory at $SUPERJEV_STATE_DIR or ~/.local/state/super-jev holds per-principal logs, the _memory config and the pointer store, plus autoheal-state and locks. It also writes prepare-cache/ and ledger/ in the checkout. The terminal app saves a pasted key to ~/.typesafe-api-key (owner-only), and install.sh writes ~/.local/bin/superjev.
python3 skills/super-jev/setup.py --uninstall removes everything Super Jev wrote: the state directory, prepare-cache, ledger, an older config that held the key, the launcher, and Claude skill links pointing into the checkout. It keeps your key file, deletes only its own names in each folder, and never touches your own files. Delete the checkout folder to remove the code.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Agent Handoff Gate: agents verify what they hand off
X: Jev as an LLM guardrail inside Agentgateway
X: Vercel's fx safety reviewer, measured at 18× faster
Reddit: Early experiments wiring Jev into Pi as a safety layer and model router.
More Jev resources like this
A TypeScript library that wraps Jev answers in a policy, a confidence gate, shadow mode, reusable recipes, and an offline eval CLI, so structured decisions can drive actions an application ships.
Jev Repos & SDKs#community#typescript#agents
Community
AntonioCoppe/jev-harnessTypeScript
AntonioCoppe/jev-harnessTypeScriptA filesystem-first framework for durable AI agents whose auto model router defaults to Jev through Vercel AI Gateway and whose evaluate helper asks typed Choice, Score, and Boolean questions inside tools.
Jev Repos & SDKs#community#typescript#agents
Community
vercel/eveTypeScript
vercel/eveTypeScriptA Juspay TypeScript SDK that treats Jev as a third inference type beside generate and stream: providers declare which kinds they serve, and tryDecide returns typed boolean, choice, or score answers, null when no decision provider is set.
Jev Repos & SDKs#community#typescript#sdk
Community
juspay/neurolinkTypeScript
juspay/neurolinkTypeScriptA Pi extension and library that gives the agent and other extensions one consented, key-managed Jev client, with a batched typesafe_evaluate tool for classify, triage, compare, and score judgments.
Jev Repos & SDKs#community#typescript#sdk
Community
DevMortimer/pi-typesafeTypeScript
DevMortimer/pi-typesafeTypeScriptAn ACP and MCP adapter that bridges TypeSafe Jev with any LLM: native Jev when TYPESAFE_API_KEY is set, or Codex, Claude, Grok, OpenCode, and OpenAI-compatible endpoints standing in as System One when it is not.
Jev Repos & SDKs#community#typescript#mcp
Community
tacticocc/JevbridgeTypeScript
tacticocc/JevbridgeTypeScript.NET SDK for the System One API with typed question sets, HttpClientFactory and dependency-injection wiring, plus Microsoft.Extensions.AI guardrail, routing, tool, and evaluator adapters.
Jev Repos & SDKs#community#sdk#csharp
Community
Hawxy/TypeSafeAI.NetC#
Hawxy/TypeSafeAI.NetC#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.