We're open sourcing our entire AI governance toolkit. Browser extension, desktop applications, SDKs, dashboard. Apache 2.0, released one component at a time over the coming months. Free to run, free to fork, free to build on, with no seat count and no expiry.
This is not a promotional gesture, and it isn't charity. It's a position, and it's worth explaining properly.
The exposure is in the wrong place
AI risk is not distributed the way security budgets are. The organisations with the most immediate exposure right now are not the banks and the insurers. They're the small teams shipping AI products at speed, with no security headcount, no governance function, and regulatory obligations arriving on a timetable they haven't looked at yet. A twelve-person company putting an agent in front of customer data has a real problem today.
Almost every tool built for that problem is priced for the enterprise that is moving more slowly and carrying less of it.
We think that gap is bad for everyone.
- Bad for the startups, who ship without controls because controls are a procurement cycle they can't afford.
- Bad for the enterprises buying from them, who inherit that exposure through the supply chain and find out about it in a questionnaire, or later.
- Bad for the credibility of the whole AI industry, which is currently arguing that it can be trusted to self-govern while pricing the means of self-governance out of reach of most of the people doing the building.
You can't audit what you can't read
There's a second reason, and it's more direct.
We ask security teams to put our software in a privileged position: between their staff and the models, between their agents and their tools, in the path of the traffic that matters most. That is an enormous amount of trust to request.
Requesting it while keeping the code closed was never going to survive a serious security review. The good CISOs push back on exactly this, and they're right to. A black box in the data path is a risk you're asking someone to accept on faith.
Now they don't have to. Read it, run it in a lab, point a red team at it, decide for yourself. If our threat model is wrong, we'd rather you found out by reading it than by trusting it.
The code was never the moat
I want to be plain about the commercial logic, because a security company saying “it's all free now” invites a reasonable question about what it's actually selling.
The code was never the defensible part. Our gateway is good engineering, but it isn't secret engineering — a competent team that set out to rebuild it could do so in a quarter. What isn't rebuildable in a quarter is everything around it: the policy library that comes from running this in regulated environments, compliance evidence an auditor will accept, the managed control plane, the support contract, someone to call when something breaks at 3am.
So the toolkit is free and stays free. If you can run it yourself, run it yourself, and we'd genuinely rather you did than paid us for something you didn't need. Prompt Shields Cloud is what you buy when running it yourself stops being the sensible use of your team's time.
The standards are being set right now
AI governance is in the window where its shape gets decided. The controls people actually run in the next two years become the default assumptions of the next ten: what gets logged, what a policy looks like, what evidence an auditor expects to see.
We'd rather that shape were set in the open, by people who can inspect and argue with it, than inside a handful of proprietary products. We're not claiming to be the ones who should define it. We're saying the definition shouldn't happen behind a licence agreement.
What's being released, and when
Component by component, over the coming months, rather than in one drop. Each release gets its own post with the threat model, the known gaps, and what we'd most like broken:
- The browser extension — prompt and data-flow visibility at the point of use, on the device.
- The desktop applications — the same controls for the AI tooling that never touches a browser.
- The SDKs — for teams putting policy enforcement inside their own agents and products.
- The dashboard — the register, the policy view, the evidence trail, self-hosted.
Apache 2.0 throughout. Everything lands at github.com/Bit-Pulse-AI, and this post is the one every release announcement will link back to.
What we're asking for
Not much, and nothing that costs money.
Run it. Break it. Open an issue when it falls over, and tell us where the threat model is wrong. If you red-team for a living and you get past the gateway, that's the most valuable thing anyone could send us. We'd rather learn it from you than from an incident report.
And if you build something on top of it, tell us. That's the part we can't do alone.