@ceyptoslim2Founder, Independent
Shipped: CUSTOS-CORE
Deterministic, fail-closed execution gate for AI agents. Every proposed action is evaluated against explicit per-tenant policy before it runs (deny by default, tool allowlists per tenant). Every decision, allow or deny, goes into a hash-chained, tamper-evident audit record. The gate is not a model, so an agent can't reason or prompt its way past it. AGPL open core.
Proof: 528 CI tests, plus red-proof CI where a deliberate one-line sabotage of the gate must turn the suite red. An independent researcher reconstructed the evaluation records, re-hashed every artifact, and published his verification: github.com/factnotebook/public-reconstructions
AI AgentsSecurityGovernanceOpen Source
Comments2
An outside researcher rechecking every record and publishing his results is proof a compliance buyer can forward to their boss. Who are you chasing for your first design partners: banks, healthcare, or another company already letting AI agents take real actions?
"Forwardable to the boss" is exactly the job that reconstruction was built for — a security team shouldn't have to take the founder's word for anything, they should be able to re-walk every record themselves. On design partners: banks last, not first. This week they're in blanket-block mode, and their procurement cycle is an 18-month conversation — they're proof the category matters, not the first customer. Healthcare is real but slow, and the compliance surface rewards incumbents. The wedge is your third option: companies already letting agents take real actions, with money movement in the blast radius. Fintech specifically — payments platforms, agent-driven finance — because a wrong agent action is denominated in dollars, the fix is measurable (deny-rate, decision record, incident replay), and "what was authorized at decision time" is the question their regulator is already asking. Honest constraint: I'm one founder pre-LLC, so I'm optimizing for one right design partner over a pipeline — someone who'll put the gate in front of a real agent workload and let the records argue.
2 comments
An outside researcher rechecking every record and publishing his results is proof a compliance buyer can forward to their boss. Who are you chasing for your first design partners: banks, healthcare, or another company already letting AI agents take real actions?
"Forwardable to the boss" is exactly the job that reconstruction was built for — a security team shouldn't have to take the founder's word for anything, they should be able to re-walk every record themselves. On design partners: banks last, not first. This week they're in blanket-block mode, and their procurement cycle is an 18-month conversation — they're proof the category matters, not the first customer. Healthcare is real but slow, and the compliance surface rewards incumbents. The wedge is your third option: companies already letting agents take real actions, with money movement in the blast radius. Fintech specifically — payments platforms, agent-driven finance — because a wrong agent action is denominated in dollars, the fix is measurable (deny-rate, decision record, incident replay), and "what was authorized at decision time" is the question their regulator is already asking. Honest constraint: I'm one founder pre-LLC, so I'm optimizing for one right design partner over a pipeline — someone who'll put the gate in front of a real agent workload and let the records argue.