siege

SIEGE: 200 people vs one agent. A typed action gate (TypeSafe System One) that learns from every breach, evaluated by W&B Weave, hardened by a defender loop. Built at CoreWeave Hacks: Agent Loops 2026.

How it uses Jev

SIEGE puts a live customer-support agent with real tools (refunds, address changes, discounts, store credit, order and profile lookups) in front of a room of 200 people whose job is to make it misbehave. Every tool call the agent proposes passes through a typed action gate, TypeSafe System One, before it can execute. Every breach is exact: the gate allowed a call, a deterministic policy oracle says it was forbidden,…

Getting started

Prerequisites: Python 3.12+, Node 20+, uv.

Source: community catalog. Details and metrics reflect the published community snapshot and are not a performance endorsement. Documentation details extracted from GitHub · the README · the project website.

Keep exploring

Similar to siege

Browse all
cyber-breach-jev project preview
Games & simulations
Cyber-Breach: The Jev Protocol - A tactical cyberpunk arena combat game powered by TypeSafe AI Jev System One decision model
jevtown project preview
Games & simulations
Jevtown: a social network where people write and 10,000 AI personas react
hundred project preview
Games & simulations
100 AI NPCs live in a tiny town. Jev chooses the next action; the world writes the story.
agent-gate-loop project preview
Agent tooling
Reusable GitHub Action: agent fix loop gated by checks, an AI reviewer, and TypeSafe Jev
rh-guard project preview
Benchmarks & research
Reward-hack radar for coding agents: structural denies + TypeSafe Jev System One sidecar for Claude Code & Cursor hooks
magic-jev project preview
Agent tooling
A Magic 8 Ball for pull requests: click the ball and it answers "should I approve this?" with one of the 20 classic phrases, chosen by Jev from the PR's real signals in about 200 ms.
pi-warden project preview
Agent tooling
Guardrails for Pi built on pi-typesafe that steer the agent instead of interrupting you: Jev judges irreversible and off-task tool calls, detects stuck loops, checks unverified done claims, flags slop
agent-fastpath project preview
Browser & computer use
Jev MCP server: a decision layer for coding agents, built on TypeSafe Jev (System One model). Ship gates, risk checks, file triage that keeps files out of context, and a safe headless browser, with calibrated confidence. For Claude Code, Codex, Cursor.
jevify project preview
Benchmarks & research
An agent skill to discover TypeSafe Jev opportunities, design typed questions, and learn from recent community experiments.