An agent audit means reading your production traces; a pilot means reading your documents. Here is exactly what access we ask for, where your data lives during the work, and what is left when we hand over.
Only what the engagement needs: scoped, read-only access to the traces, logs or documents in scope, granted by your team and revoked at handover. No standing admin access to production.
In your environment wherever possible: your cloud, your repo, your CI. Where samples leave it, they are de-identified first, and nothing goes to a third-party platform you have not already chosen.
Our working copies of your data are deleted at handover. What remains is in your repo: the code, the golden set and the documentation.
Signed before any data or code is shared. Yours or ours, whichever your legal team prefers.
Everything we build for you is assigned to you, with full IP transfer on final payment. No licence back, no platform lock-in.
For health, financial or personal records, we work on de-identified samples or entirely inside your environment, and agree the DPA or BAA terms your policy requires before kickoff.
Yes. An NDA is in place before anything is shared, and we are happy to sign yours rather than ours.
Send us your DPA and we review it before kickoff. If your data includes health or other regulated records, raise it on the first call: we scope the work around de-identified samples or inside your environment, and agree the paperwork your policy needs before any data moves.
No. Your traces, documents and labels are used for your engagement only. They are not used to train models and not shared with other clients.
Not unless you already use one. The harness runs in your repo and your CI; if you run LangSmith, Braintrust or Langfuse, it plugs into your account.
You do. Code, golden sets, judge prompts and documentation are assigned to you, with full IP transfer on final payment.
Sending a security questionnaire? Email it to contact@rottawhite.com and we return it before the scoping call. How the work itself is checked is on the eval harness and audit trail pages.
Next step
Audit the agent you already run, or pilot one workflow on your own documents. Either way, you leave with a number you can check.