Unlocking context and data for your AI agents.
Boston, MA
Joined November 2020
- Tweets239
- Following40
- Followers80
- Likes15
hoop.dev retweeted
I got some agents running in the background with my "things I'd do if I had time" list.
And now we are about to put some tests in a thing that helps you use @hoopdotdev with your agents to protect things at protocol level
shhhh, Mat is cooking 🤫
I got some agents running in the background with my "things I'd do if I had time" list.
And now we are about to put some tests in a thing that helps you use @hoopdotdev with your agents to protect things at protocol level
hoop.dev retweeted
Small thing that makes me happy: the agent writes a bad DELETE, hoop.dev blocks it, the agent reads the error and hands it to a human. Nobody had to be watching.
hoop.dev retweeted
Jev is a classifier, and the hype around it says one thing: guardrails for AI agents stopped being a hypothesis.
Most tools put the check inside the agent. We put it between the agent and the database. Testing Jev there this week. 👉 @hoopdotdev
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Most teams put their agents through the same gates as humans. Open a PR, get a review, let CI apply it. But if you want an agent to produce at human speed, why use agents at all?
Autonomous agents need autonomous controls. We're building an open-source proxy between agents and prod, and the review happens at runtime. At machine speed.
If you've handed an AI agent database credentials: what worries you more, the access itself or what happens after?
We asked in 2 dev comminities this week. People had strong answers for the access. Almost nobody had one for the after.
What we're building for the after, for free: hoop.dev/start
Today at 1:00 PM EDT: Runtime Executive x hoop.dev join forces to discuss how People, Process and Technology combine to control agentic AI.
Register here: luma.com/7iusobve
Tomorrow, 1 PM ET: how do you know AI did what you intended? AI governance that enforces while the agent acts.
With Greg Crowley (former CISO at WWE & eSentire) and Andrios Robert, our founder.
hoop.dev/webinar/ai-accounta…
Tomorrow, 1 PM ET: how do you know AI did what you intended? AI governance that enforces while the agent acts.
With Greg Crowley (former CISO at WWE & eSentire) and Andrios Robert, our founder.
hoop.dev/webinar/ai-accounta…
A static rule can't judge intent. A human can't review at machine speed. The only thing fast enough to stop an AI agent is another agent, reading every query live. We built that agent. Free and open source: hoop.dev/start
There's no screen in AWS that shows you everything you're running. The bill you pay show only one number.
Let blueprint help: one command, read-only, and it breakdown the count for you.
Easy install, btw: github.com/hoophq/blueprint
We just bundled a cool tool set in one plugin for Claude Code. A guardrails for destructive commands, a risk scanner for personal data/secrets in your sessions, and a token saver.
Built for solo builders: github.com/hoophq/claude-mar…
You pay for everything your coding agent does, even though most of it is unnecessary for what you need.
Julius, the newest free launch from our eng team, filters command output before it reaches the model.
Verbose commands compress 60-90%, and errors always pass through.
Exact per-app, per-model token metering.
Here's the Repo: github.com/hoophq/julius
Your agent doesn't need your real API keys to do its job.
Cloak, our new standalone free tool, hands it a fake credential and swaps in the real one only the way out.
The real secrets live in your OS keychain, but never in context or logs.
Check it now: hoop.dev/labs/credential-clo…
PII detection for Go, with no service to deploy.
Alcatraz detects 45 entity types across 12 countries in-process: no network, no models to download, and the core is pure standard library.
10 to 100x faster than Presidio in our benchmarks, and it's a standalone open source library.
Free for everyone: github.com/hoophq/alcatraz
Your AI coding agent's session logs hold more than you think: pasted configs, keys, customer data inside stack traces. hoopRS, our new free open source tool, scans them right on your machine and shows you what's there. Your session content never leaves your environment.
MIT licensed. No signup wall and no trial. Runs locally. Repo is github.com/hoophq/rs