Start with the audit. The evidence is yours either way.

The entry is the Agent Exposure Audit: Interpose in transparent mode on your Claude Code fleet for two weeks, ending with the Rollout Evidence Report, the signed record of what your agents actually did and what policy would have caught. Fixed fee, agreed in writing before we start. Deployed by Tuent inside your boundary.

Scope the audit Read the brief first

Three rungs. Each one earns the next.

You never buy the platform to find out whether you need it. The audit shows you what your agents do. The pilot proves the controls on your own work. The subscription runs the fleet.

sessions

1,204

tool calls

318

files read

egress targets

pattern breaks

j.reyes	read	.env.production during a formatting taskoutside this role's usual scope	surfaced
m.osei	net	egress to an unlisted host, once, mid sessionnot seen before this week	surfaced
a.khan	read	41 denials against the credential floorrepeated reach for the same secret	surfaced

rollout evidence report · payments-api · illustrativenothing blocked except credentials

The audit. Interpose watches, your developers change nothing.

In transparent mode Interpose records and takes no action, with one exception: credentials and keys are blocked from day one on every tier, and a developer sees that as a normal denial. Everything else is written to the signed trail while your agents keep working. At the end you get the Rollout Evidence Report: the tools called, the files read, the hosts reached, what would have been blocked and why, and anything that broke the agent's own pattern.

It is usually the first complete picture a team has of what its agents do.
The report is yours to keep, whether or not you continue.

The pilot. Each finding becomes a rule your team approves. Illustrative.

finding	read	.env.production during a formatting task	becomes
rule		credentials and keys	block
finding	net	egress to an unlisted host	becomes
rule		network allowlist for this role	block
finding	read	41 reaches for the same secret, auth-service	becomes
rule		files outside the role	alert
rule		destructive shell commands	ask

Interpose drafts, your team approvesthen block mode, one control at a time

The pilot. The report becomes policy, then policy blocks.

We sit down with your team and turn each finding into a decision: watch it, or stop it. Interpose drafts the rule, your team approves it, and each control moves from alert to block one at a time. In week three, block mode runs live on the use case and your team watches Interpose stop real behavior on real work.

Nothing tightens without a human decision.
When a block is wrong, the developer releases it in one click and the rule tightens from the miss.

The shape of the pilot.

Terms agreed before it starts, with a founder on the other end throughout.

Who it is for: A team running agents in production on Claude Code, with a named owner who can say yes. Claude Code is the live integration today.
Where it runs: On your machines, under your keys. Tuent installs the hook with your team on a call, under your credentials, and holds no access afterward. Nothing about your agents, code, or actions reaches Tuent.
What you get: The Rollout Evidence Report, the policy your team approved, and the full signed record of what was watched and what was stopped. Yours whether or not you continue.
What it costs: The audit is a fixed fee, agreed in writing before we start. The pilot is agreed the same way. If you continue, a letter of intent signed during the pilot states the annual rate for your fleet size, and that number does not move as the fleet grows.
What we ask of you: One use case, one owner, and a few hours of your team's time across the three weeks for scoping, the week one review, and the close, plus whatever your own security review needs before an install.
If you stop: The hook is removed on the same kind of call it was installed on. You keep the report, the policy, and the trail.

Thirty minutes to scope the audit. Then the weeks you can watch.

We map where your agents run, agree the fixed fee in writing, and open the audit in transparent mode. The report is yours whatever you decide next. If the brief would help first, it is a seven minute read.

Scope the audit Get the technical brief

Or write to james@tuent.ai or charlie@tuent.ai. A founder answers.