Orivael

The control plane for AI agents — govern before they act

8 followers

The control plane for AI agents — govern before they act

8 followers

Orivael sits between your users, your agents, and your tools and enforces policy on every action before it runs. It blocks unsafe or irreversible tool calls, routes high-risk actions to human approval, and signs every decision into a tamper-evident audit trail. Built for teams shipping AI agents that can't afford a "the bot deleted prod" incident. • Intercepts tool calls, not just text • PASS / WARN / BLOCK + approval • Signed audit log • Self-hostable, microsecond hot path

This is the 2nd launch from Orivael. View more

NodeXLoop — Agent Loop DevTools

Launching today

Debug AI agents like you debug code

A visual debugger for AI agent loops, inside VS Code. Build Agent → tool → critic → retry → gate → test → approve on a canvas, run it node-by-node with breakpoints, and see exactly which call the guard blocked and why. The graph is a .loop.yaml next to your code, so it reviews in the same PR. Axiom guard on by default: a stalled critic loop drops from 20 model calls to 6, a prompt injection costs 0. 37 templates, org review (local preview), audit export. Free on the Marketplace and Open VSX

NodeXLoop — Agent Loop DevTools gallery image

Free

Launch tags:Productivity•Developer Tools•Artificial Intelligence

Launch Team / Built With

LightfieldAI-native CRM that builds itself and does work for you

Promoted

Maker

📌

Hello PH :) Most agent tooling looks like n8n: Salesforce → Gmail → Slack. That's the wrong shape for what agent loops actually look like — Agent → tool → critic → retry → gate → code change → test → approve. NodeXLoop is built for that shape specifically, as a VS Code extension that lives next to your code instead of a separate dashboard. The part I'd genuinely like feedback on: every graph runs behind an Axiom guard by default, which rules PASS/WARN/BLOCK on every model call before it moves. Measured against the built app, not claimed: a critic loop that's stopped improving goes from 20 model calls to 6 with the guard on; a prompt-injection input costs 0 calls; a loop that's genuinely still improving is untouched (12 calls either way). I'm not going to pretend it's flawless — per-node runtime permissions are currently recorded and audited, not yet enforced by the runtime, and that's stated plainly in the README rather than glossed over. Free, install from the Marketplace or Open VSX, or grab the .vsix directly. Would love your hardest questions — especially "what happens when the guard is wrong."

Report

15h ago

Previous Orivael Launches

OrivaelThe control plane for AI agents — govern before they act

Launched on July 7th, 2026