Your customer's security review wants more than your word.
Companies buying an agent that will touch their money or their customers' records ask what it can do, what it's refused, and how they'd know. The Ancilla flight recorder gives them a packet they can check for themselves: who instructed the agent, what it was allowed to do, and what actually crossed the gateway.
When something goes wrong, the argument is about whose fault it was.
When an agent causes harm, the business may blame the developer, and the developer may answer that the customer misconfigured the agent or gave it too much access. Without a record both sides trust, that's one account against another.
Who gave the instruction?
The instruction is recorded with the identity it came from, so "the agent decided on its own" and "the customer told it to" can be told apart.
What was it allowed to do?
The customer's grant is signed and on the record: which actions, which accounts, what limits. If access was too broad, that shows too.
What it actually did
What the gateway forwarded and what the target system reports, beside what the agent claims. Where they differ, the verifier says so.
On disputes between businesses and developers: Insurance Business, July 2026.What your customer runs
Ancilla's decision service and recorder, inside their boundary, under a decision key only they hold
Their own gateway, which carries out each decision. We have run this with Apache APISIX; other gateways need integration work
What Forerunner Labs holds
Signed, dated checkpoints of each record, never its contents
Your customer's data stays with your customer. That is a design we have built and tested ourselves; no outside party has tested it yet.