Skip to content

Permissions and trust

A connected assistant can do what you can do, and nothing more. It signs in as you, it is limited to your own data unless you ask about your team, and you can cut it off at any time.

This page explains what the assistant can read, what it can change, what it can never reach, and how to end access. If you are reviewing Vaam for your security team, start here.

By default, every tool returns your own data only. When your assistant lists sequences, prospects, videos, tasks, or issues, the list contains your records and no one else’s, even when you share a team.

Your team’s data is included only when you ask for it. Four kinds of list can widen to the whole team, and only on request:

Ask aboutAssistant can widen to the team
SequencesYes
ProspectsYes
Prospects’ progress in sequencesYes
Issues affecting outreachYes
VideosNo. Always yours only.
TasksNo. Always yours only.
Account totals and limitsNo. Always yours only.

Team lists also require Read team information. Ask explicitly about the team, such as “How many prospects is my team working right now?” A question about your numbers stays about your numbers.

Single-record reads can open a teammate’s record only where that tool supports team access and you grant Read team information. Prospecting agents and reply lists require ownership. The assistant cannot open another team’s data through the customer address.

Technical clients must not send userId to the customer address. Any userId key returns invalid_arguments before tool execution. The rejected tool performs no data reads, writes, or external calls. The connection determines your identity.

Writing is a separate set of permissions from reading, and you see the full list before you approve. A connected assistant can:

  • Pause and resume a sequence.
  • Change the subject, body, or video of a sequence step’s message. This never changes a sequence’s delays, step order, or step types.
  • Add prospects to a sequence, correct a prospect’s contact details, and remove a prospect.
  • Pause, resume, mark as completed, or permanently stop outreach to one person.
  • Mark an issue as resolved or dismissed.
  • Create a video and finish uploading it.
  • Change your AI outreach settings: your ideal customer profile, company details, offering, and message language.

If you are a team owner or team admin, changing the AI outreach settings also changes them for every member of your team. That is how those settings work in Vaam generally. The assistant follows the same rule.

If you approved Write sequences before message editing launched, your assistant already has it. Permissions on the approval screen cover a category of action, not a fixed list of tools. Before message editing launched, the approval screen called this permission Write sequences. It now reads Pause and resume a sequence, and edit its messages. An assistant your account already trusts with it gains the ability to edit a step’s message the moment Vaam ships it, with no new approval screen. If you would rather review this specific ability first, remove the connector and reconnect it.

A step’s message can also change and quietly drop its video: if the new text no longer includes the tag that shows a video, Vaam removes the video from that step rather than leaving a tag-less video configured. Your assistant should tell you when this happens.

An assistant never sends outreach itself. It starts and stops the work; Vaam performs it, on your normal schedule and within your normal limits.

Tools identify read operations and changes. They also identify changes that overwrite or remove data. Your assistant should explain the effect and obtain your agreement before a change. Reads do not need that confirmation. These instructions do not add a separate Vaam confirmation screen to every action.

If a change fails or times out, check the current state before you repeat it. Server logs provide limited evidence. They do not guarantee a durable audit trail or exactly-once external effects. A failed response alone does not prove that no message or external update occurred.

No matter what you approve, a connected assistant cannot:

  • See data belonging to another team or to a Vaam customer who is not you.
  • See your password. You type it into Vaam’s own login page, in your own browser, never into the assistant.
  • Reach billing, invoices, or payment details.
  • Reach Vaam’s internal support and administration tools. Those live on a separate address, described below.
  • Do anything after you disconnect it.
  • Grant itself more than you approved. You can narrow permissions later, but you cannot widen them. Vaam checks them against your live account rather than freezing them when you approve, so the connection follows changes to your role in Vaam.

The approval screen lists permissions in plain terms. This is what each one covers:

On the approval screenWhat it allows
Read sequencesSee your sequences, their steps, and their stats
Pause and resume a sequence, and edit its messagesPause and resume a sequence, and change a step’s message
Read prospectsSee your prospects and their details
Write prospectsAdd prospects to a sequence, correct their details, remove them
Read enrollmentsSee sequence progress, retained status changes, and retained reply evidence
Write enrollmentsPause, resume, complete, or stop outreach to one person
Read videosSee your video recordings and their view stats
Create videosUpload a new video
Read statsSee your counts, totals, and account limits
Read work itemsSee issues affecting your outreach
Write work itemsMark an issue resolved or dismissed
Read user informationSee your name, email, and account type
Read team informationSee your team’s name, size, and connected integrations
Read settingsSee your AI outreach settings, prospecting agents, and their sources
Write settingsChange your AI outreach settings

The progress permissions cover each prospect’s sequence progress. Reading retained history and replies uses Read enrollments. Reading agent qualification results needs both Read settings and Read prospects. Reading tasks needs both Read enrollments and Read prospects.

If you approve without the assistant asking for anything specific, you grant this whole list. An assistant that asks for a shorter list gets only what it asked for.

An API key carries the same permissions, except that you tick them yourself when you create the key. A key that only needs to read can be created with no write permissions at all. You cannot add permissions to a key later. Create a new key instead.

You can end access at any time:

  • Remove the connector in your assistant. See Disconnecting for the steps in each app. This deletes the tokens the assistant was holding, so it cannot sign in again.
  • Delete the API key under Settings → API. The key stops working immediately.
  • Email support@vaam.io if you cannot reach the assistant or think a connection has been misused. Vaam can cut it off.

For a security review, note two details. A renewal token is single-use. Vaam replaces it every time it is used. If an old one is presented again, Vaam revokes the whole connection immediately. An access token expires one hour after it is issued, which limits how long a token already in flight can keep working after access ends.

Nothing in your Vaam account changes when access ends. Anything the assistant already did stays done. A paused sequence stays paused until you resume it. You can see what it did in your normal sequence and prospect history, because an action taken through an assistant is recorded the same way as an action taken in the Vaam app.

Vaam runs a second MCP address for its own support staff, at a different URL. Customers cannot use this address with a customer connection.

Access granted for the customer address fails at the administration address. The reverse also applies. Customer approval cannot grant access to staff tools.

The staff address has the same 34 shared tools and 14 legacy admin-only tools. All 13 shared changes are available there under the same names. Each shared call requires userId to select the target account.

Shared reads require a live staff admin role, admin:read, and the tool’s domain permissions. Shared changes require a live staff admin role, admin:write, and the tool’s domain permissions. Team requests also require team:read. An admin permission alone does not grant domain access. These checks also apply to shared help-document calls, which require a valid target user.

Shared operations run with the target account’s user role, team settings, ownership checks, and limits. They do not use the staff member’s account settings or unrestricted admin access. An immutable actor identity records which staff member acts. The target remains the account whose data changes.

Before a targeted change, the assistant should confirm the account and effect with the staff member. Adding prospects or resuming outreach can start messages for that account. Changing AI settings follows the target’s team role. Existing consent permissions still apply. Shared tools add no approval screen for every action or record of every read.

Vaam never sends your data to Claude, ChatGPT, or any other assistant on its own. It travels there because you connected the assistant, and only in answer to the question you asked. This is the same path as opening the Vaam app and copying the answer into a chat yourself.

Once it arrives, the assistant’s provider handles that data under its terms, not Vaam’s. Anthropic handles Claude data, and OpenAI handles ChatGPT data. Choose assistants with the same care you would apply to any other tool that reads your customer data. Check your provider’s data retention and training settings. See the AI assistants section of Vaam’s Privacy Policy for Vaam’s own position.

Vaam’s handling of your data is unchanged by connecting an assistant. See Personal Data Management, GDPR, and Sub-processors.

Questions about a connected assistant, or about anything it did on your behalf: email support@vaam.io.

For Vaam’s full data handling commitments, see the Privacy Policy and the Terms of Service.