Trust
We read, we never change. You see every permission before granting it, you can switch any of them off, and revoking removes the whole thing — the app included.
No. Every permission we ask for is read-only. The one exception proves the rule: the app holds a single write permission that lets it remove itself from your Microsoft account when you revoke access — it can delete itself, and nothing else.
Each assessment reads a slice of how your business runs — who has access to what, what you're paying for, where the work actually flows — and hands back a written, ranked report. You choose the assessment, you approve what it may read, and the report is the product. We don't change anything, and we don't do the fixing unless you separately ask us to.
For assessments that read your Microsoft 365 or Azure account, someone with admin rights has to approve the access — that's a protection, not a hurdle. If your IT is a contractor, a short call with them is usually all it takes. Assessments that work from documents you hand us need nobody.
No — and you can verify this independently rather than take our word. We use Azure OpenAI, a private deployment inside Microsoft's cloud. Microsoft's terms state that prompts and data are not available to other customers and are not used to improve any models. This is a different arrangement from the public ChatGPT.
The contents of anyone's email (headers only — subject, recipients, timestamps). The contents of Teams messages (structure only — which teams and channels exist). The contents of files (names and sharing settings, never what's inside). Passwords, secrets, and tokens — never. The permission model makes reading this content impossible for the app, not merely against our policy.
Aggregate findings — the shape of things, like what fraction of accounts have a second sign-in factor, or how many policies apply to everyone. Plus the business context you choose to give us. Individual user records, message contents, and file contents are never sent, because we never hold them in the first place.
In Microsoft Azure (East US), encrypted at rest and in transit, with credentials held in Azure Key Vault. The results are yours: reports are visible only to the people in your workspace, and we never share them with anyone else. What you do with the findings is entirely your decision.
Yes, yourself, without asking us. Any workspace admin can reset or remove an assessment from the portal, which permanently deletes the report files, the intermediate data, and the database records. There is no minimum retention period and nothing is locked in.
Every one of them is listed below, generated live from the app's own configuration — so the list can't drift from reality. A small required set covers basic profiling; everything else is optional.
Yes, individually, in Settings, before anything runs. A disabled permission means the corresponding check simply doesn't happen — no reading, no collection — and the report says plainly what was skipped rather than papering over the gap.
One button in Settings revokes access, and the app removes itself from your Microsoft account — that's what its single write permission is for. If you'd rather do it from your side: in your Microsoft admin center, find "Agent Fishbowl - Environment Auditor" under Enterprise Applications and delete it. Either way, access ends immediately and any cached access expires within the hour.
None yet, and we'd rather say so than decorate this page with the certifications of the cloud we run on. What we can show you today is better than a badge: the exact list of what we can read (above), the fact that all of it is read-only, and a revocation that removes the app entirely. When we hold a certification of our own, it will be named here — and not a day before.
Only the people working on your engagement. We're a small firm — the person accountable for your assessment is the person you can email about it. Infrastructure access exists to run the service; nobody browses client data without cause.
Ask it. support@agentfishbowl.com is read by a person, and hard questions about access and data are exactly the ones we want before you grant anything.