AI/ML●●●Banger
Decompose – Split text into classified semantic units, no LLM, 14ms
Non-LLM deterministic semantic decomposition—14ms, no hallucination, MCP-ready.
Big BrainSolve My ProblemWizardry
echology-io
103mo ago
Back to browse
GitHub Repository
One gateway in front of every protocol. Same policy across MCP, LLMs, databases and containers. Wire-level enforcement at under 5ms.
721 starsGo
We're using LLMs to classify risk before execution in prod
by andriosr·May 5, 2026·2 points·0 comments
AI Analysis
●●●BangerSolve My ProblemSlick
Wire-protocol parsing blocks DROP TABLE before execution with zero code changes.
Strengths
- •ML-based PII detection masks sensitive fields automatically without regex configuration
- •Human approval workflows via Slack/Teams for risky operations before they execute
- •Session recording with full replay for compliance audits and incident investigation
Weaknesses
- •Agent governance features depend on teams actually deploying AI agents in prod
- •CNCF membership and SOC 2 suggest enterprise focus over indie developer appeal
Category
Target Audience
DevOps, SRE, and security teams managing production infrastructure access
Similar To
Teleport · StrongDM · Tailscale
Post Description
Hey HN!
I'm Andrios, founder of Hoop.dev, an OSS layer-7 gateway for infra access. We just released a new integration: put LLMs between devs' or agents' actions and databases or Kubernetes.
The model gives a more nuanced analysis of the action, not only the syntax. Like: is this deleting data? Updating large number of entries? How risky is it? Then it decides if it will allow the execution, send it to human approval, or block it completely.
Because Hoop sits at the network layer, this process happens in-transit, as data is passing through the gateway. Very low setup required.
As product teams put agents in production, we're seeing security and SRE teams also shipping agents to enforce controls, and this is a nice way of deploying them.
What do you think about this approach? Any feedback is super welcome here.
Project is here: https://github.com/hoophq/hoop
Similar Projects
AI/ML●●Solid
Using a mobile LLM app to safely operate a desktop computer
Separates LLM cognition from OS execution via discrete actions, not continuous control.
Big BrainBold Bet
Ruikhu
123mo ago
Developer Tools●Mid
Guard – An open-core governance layer for AI-generated code
Governance for AI code is real problem, but Guard is vapor—roadmap, not shipping product.
Bold BetBig Brain
veeduzyl
203mo ago
AI/ML●●●Banger
Castor – a secure execution layer for LLM agents
Kernel interception stops runaway agents where LangGraph and AutoGen only advise.
Big BrainSolve My ProblemShip It
claytonia
102mo ago
AI/ML●●Solid
Runtime governance layer that refuses high-risk LLM outputs
The demo implements post-generation admissibility checks and returns structured refusals (decision codes, rule triggered, divergence metrics and a stable prompt fingerprint) so you can audit enforcement decisions. It's a crisp, focused proof-of-concept for runtime enforcement — useful as a starting pattern — but it stops short of addressing bypass/adversarial vectors, deployment integration, or guarantees that make it enforceable at scale.
Niche GemShip It
milarien
113mo ago
Infrastructure●●Solid
LLM-Gateway – Zero-Trust LLM Gateway
Zero-trust networking via zrok beats LiteLLM when your GPUs sit behind NAT.
Big BrainSolve My Problem
michaelquigley
712mo ago