We've been building things. Tools to fix problems AI rollout at scale created. A development methodology that came out of the gap between what AI coding tools promise and what enterprises can actually deploy. Architecture patterns we keep rediscovering across client engagements. The kind of work that's too technical for an executive audience, too opinionated for a case study, and too useful to leave inside the building.
That work needs a home. So we're giving it one.
Intevity Labs is a new section of the site, sitting alongside Insights, for the technical work coming out of Intevity. Open-source tools, architecture writeups, methodology pieces, and engineering points of view. Things our team has built or learned that we think are worth sharing under their own names, with the level of detail engineers actually want.
What you’ll find here
Three kinds of things, roughly:
- Tools — open-source apps and libraries you can install and run today.
- Applied research — experiments and prototypes that explore where the work is heading, with the rough edges left visible.
- Field notes — engineering write-ups on what we shipped, what broke, and what we’d do differently.
Why we’re doing this in the open
Because the best way to prove a tool works is to hand it to someone else and watch them use it. Open source keeps us honest: the docs have to be real, the defaults have to be sane, and the thing has to stand on its own outside the context it was born in.
The best way to prove a tool works is to hand it to someone else and watch them use it.
Starting with Sentinel
Our first release is Sentinel — a desktop companion for Claude Code and the Claude desktop app. It runs as a small tray app with a bundled local proxy, so every request to Claude passes through your own machine, where Sentinel can inspect it, route it, and keep an eye on cost. Three things we kept wanting on every AI-assisted build:
- In-flight security scanning — it catches secrets, PII, prompt injection, and risky tool calls before they reach the model, in observe or block mode.
- Multi-account routing — switch between the Claude accounts you own, or let Auto mode fall back to whichever still has capacity so you’re never stuck mid-task.
- Cost optimization — cheaper subagents, reversible payload compression, and usage metrics that show where the tokens actually go.
There’s no package-manager step and nothing to wire up by hand. Download the build for your platform from the releases page and you’re a few clicks from running:
- Download & launch — grab the
.dmg(macOS),.msi(Windows), or.deb/.rpm/.AppImage(Linux). The tray icon appears and the local daemon starts on its own. - Activate — click Activate Sentinel and restart Claude Code so its traffic runs through the proxy.
- Add an account — complete the one-time OAuth for each Claude subscription you want to route.
That’s the whole idea of Labs: take the thing that quietly made our own work better, and make it easy for you to pick up too. Sentinel is open source under the MIT license — the code, docs, and issues all live on GitHub. More soon.