Hey, this is Ben. I was the lead AI researcher on this platform in its early days. The team fallout last year was difficult, but I also think there’s a lot we can learn from it.
I did most of the AI work behind the platform, and with the original founders having moved on to pursue different things, I feel a strong sense of responsibility to carry the work forward and see what we can build from here.
I’ve taken over the platform and decided to move forward with a much more challenging mission—one that I believe could help define where AI is headed next.
This update makes it easier for teams to understand:
- when a request is stopped by policy
- how unsafe inputs are rewritten or contained
- where explicit authorization is required before inference
- how the private execution path behaves end to end
Quick start:
1. Start server: npm run server
2. Health check: GET /api/firewall/health
3. Gate model calls through /guard
4. If blocked, return safe fallback to user
Honestly, the hardest part about plugging AI agents into a root-level system isn’t getting them to work. It’s dealing with the damage they might cause.