You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$200,000 left to reach the funding goal.
AgenTrust is an open source project that governs AI agents. It gives you a way to cryptographically prove what an agent actually did, which tools it called, under what policy, with what authority, and whether it stuck to the rules it was supposed to follow. You never have to look inside the model itself to check any of this.
Right now the stack is eight repositories. There’s a signed record format for what an agent did, a gateway that checks policy on every tool call as it happens, a way for agents to hand work off to each other without over-granting authority, a public log anyone can check for themselves, and a system that protects a model’s weights before it’s even deployed. Code is athttps://github.com/agentrust-io/
Earlier this year the Linux Foundation’s Agentic AI Foundation gave AgenTrust its Open Agentic AI Innovation Award. They credited it with close to half a million downloads in its first three months, and 144,000 just in the last 30 days.
Get governance down to one line of code. Right now, wiring the five pieces together to actually use this takes a full day. That should be a decorator, not a day.
Build a dashboard that puts policy decisions, blocked actions and verification results in front of the people actually running these systems, instead of leaving it buried in a log somewhere nobody reads it.
Make the pieces check themselves against each other continuously, not just once at startup. A deployment should always be verified against what it claims to be, the whole time it’s running, not just the first minute.
Support every major way people are building agents right now, from Claude Desktop to whatever no-code platform shows up next, so this works wherever an agent actually lives.
We’ll get there the same way we’ve gotten this far: building in the open, with the contributors who are already shipping alongside us, and getting it in front of real outside teams early so the roadmap is shaped by what they need instead of what we’re guessing they need.
Real outside teams running this in an actual deployment, not people who starred the repo and moved on. A pass/fail test any standard AI client can run against the gateway and actually pass, which it can’t do today. An independent security review of the new pieces, especially the dashboard and the integration layer, that comes back clean before we launch.
Mostly covering living costs, so this becomes full-time work instead of evenings and weekends squeezed in around something else. Some of it on compute and API costs, since testing across multiple model providers in real deployments isn’t free. The rest on a proper security review, because the dashboard especially is going to be handling sensitive governance data and that’s not something to cut corners on.
I’m part of the core team building AgenTrust. I maintain all eight repos and review what goes out to PyPI. A real community has grown around it since then, a solid group of outside contributors with merged PRs across our repos, and a few of them have taken on reviewer and maintainer roles themselves now, shipping changes we didn’t write.
The most likely way this fails is we build the wrapper and the dashboard well, and adoption stays low anyway, because the real complexity, writing an actual policy, running an actual gateway, is still sitting there underneath. If that happens, we’d rather end up with one or two deployments we really supported than a lot of shallow coverage nobody actually depends on. Whichever of the two pieces shows real outside use first is where the rest of the funding goes.
Nothing.
There are no bids on this project.