Inline Sign in
SECURITY & GOVERNANCE

The question is not what it can do. It's what it can't.

Handing an assistant an API key gives it everything that key can reach, forever, with no record. Every control below exists because that is the thing being replaced.

CREDENTIALS

The assistant never holds a key.

Sources are connected once and stored by Inline. An assistant gets a tool it may call, not the token underneath it. Disconnect a source and every tool built on it stops working immediately, and we then try to revoke the token at the provider too — which many providers do not allow, and we tell you when that is so. Inline itself can read what it stores, because it has to use it on your behalf. The key that seals a stored credential depends on your plan: some plans have a key of their own and the rest share one. What a key of your own does and does not cover is on the pricing page, and it holds only within the limits on that page.

SCOPE

It can only reach what you ticked.

An API description with 312 operations becomes the four you published. The rest are not hidden or discouraged — they were never registered as tools, so there is nothing to call.

APPROVAL

Some calls stop and wait for a person.

Mark a tool as needing a human and it will not go through unattended. The call waits in Approvals with its arguments visible, and whoever the rule names releases it or does not.

AUDIT

Every tool call is written down.

Who ran it, against which account, with what arguments, how it ended, and how long it took. Arguments have credential-looking values removed. What came back is not kept by default — only its size and a fingerprint of it — unless you switch that on for a source. Sessions are visible while they are still running, not reconstructed afterwards.

PEOPLE

Access is a group, not a favour.

Put people in groups and give each group its own servers and rules. Two-step verification is available to every person, who turns it on for themselves. Requiring it of a workspace, and sign-in with your company’s identity provider, are set up with us: for the provider we create it, and then you choose which domains must use it.

BLAST RADIUS

Installing someone else's process runs it on your accounts.

A published process is a shape, not a key. When you install one it runs against the sources your workspace has connected — so before it runs, you are shown exactly which of your systems it would touch.

WHAT A RECORD LOOKS LIKE

Not “the assistant did something”.

An audit line is only useful if it answers the question somebody will actually ask six weeks later, which is never “was there activity”. It is who, on whose authority, against which account, and what changed.

  • Readable by the person who ran it, not only by an admin
  • Kept for 365 days, on every plan
  • Visible live, while the session is still going
  • Held and refused calls are recorded too, not just successful ones
14:22:07  maya@   new_client_onboarding
          client Northwind Ltd

  stripe    billed    standard      ok
  asana     opened    ONB-4471      ok
  gmail     sent      welcome       ok

14:24:51  sam@    delete_order
          HELD - waiting on finance

14:25:14  sam@    export_all_users
          REFUSED - not a published tool
WHAT WE DO NOT CLAIM

Things worth asking us about.

Inline is early. We would rather you found the gaps on this page than in a procurement call. We hold credentials for the systems you connect, which is a real responsibility, and our servers can read them, because they have to use them. We do not have a SOC 2 report, an ISO 27001 certificate or a formal pen-test to point at. Workspaces are separated in software, not in separate infrastructure. Deleting a workspace does not delete everything: billing records and the audit trail are kept on purpose, and backups hold copies for up to 14 weeks. One path is not recorded: when an agent host asks a connected service for one of its app panels, Inline forwards the request with your credential and does not write a call record for it. If your review needs any of this to be different, talk to us before you build on it rather than after.

Exactly what is stored, for how long, who else receives it and what deletion reaches is in the privacy policy, which is still a draft that has not been through a lawyer.

Try it free See how it works