The company
Kyance is building trust infrastructure for agentic commerce: identity verification, customer profiling and real-time trust scoring for merchants whose buyers are increasingly agents rather than people.
It handles identity data at scale, and it sells to companies whose procurement teams ask for a SOC 2 report before they ask about anything else. Tim Brown builds the platform. There is no Head of Security, no GRC lead and no DPO.
The challenge
Tim had spent his career in identity. He knew precisely how a SOC 2 audit goes for a team that leaves it until a customer demands one.
Everyone I know who has been through SOC 2 after the fact describes the same thing. Months of hunting down evidence for decisions nobody wrote down at the time.
The harder problem was pace. Kyance changes every week, much of it written with a coding agent. Nothing flagged which of those changes carried a SOC 2 consequence.
I'm building this platform largely on my own. If I change something, or I have Claude change something, there was no way for me to know whether it just broke a control.
The solution
Tom Giambelli scoped the SOC 2 system description against the commitments Kyance actually makes to its merchants, then set the trust services criteria that follow from them: security, availability, confidentiality, processing integrity and privacy. Scoped to what Kyance claims, not to a template.
The risk register was built from the real architecture. Thirteen scenarios, each mapped to the controls that reduce it, each sized for a two-person company rather than an enterprise, so nothing was implemented that the business did not need.
Then the Keldyn MCP server connected into Claude Code, the environment Tim already builds in. He asks which SOC 2 controls his codebase satisfies and what still needs work, and the answer arrives in the same window as the code. Keldyn never takes a copy of it.
The first time I ran it, it went through the whole codebase, told me what was missing, and pushed the lot into Jira as tickets. It did a serious amount of my work for me.
GitHub was connected alongside it, so configuration evidence collects continuously rather than being assembled the fortnight before an auditor arrives.
The impact
SOC 2 findings now arrive as Jira tickets under a dedicated epic, prioritised alongside product work rather than competing with it. When Tim refactored the platform to handle protocol changes, he reran the assessment and confirmed in minutes that nothing had regressed.
Kyance will walk into its SOC 2 audit with the evidence already gathered. The deeper change is the order of operations.


