Skip to content
BTS 2026Met us at the summit?Claim your summit offer
Limits an admin sets for every AI agent

An Agent Can Only Do What Its Mandate Says

Connect Claude, ChatGPT, Cursor or your own agent. An admin issues each one a mandate: which templates, which actions, which recipients, how many sends a day, and until when. The limits are enforced on FlowVerify’s servers, so whatever the app does, it cannot change its mandate, and a request outside it is refused before anything is certified or emailed.

  • Refusals name the limit that was met
  • An agent never signs for anyone
  • Certificate names the agent, mandate and issuer (IT Act, s. 11(c))

What a mandate holds

The Settings That Decide What an Agent May Do

An admin issues a mandate in the Agents console. It is not a secret and it grants nothing by itself: the agent still needs its API key or its signed-in connection, and the mandate narrows what that may be used for.

Agent label
The name on every audit row the agent causes, and on the certificate of completion: “sent by agent … under mandate …, issued by …”.
API key
Bind the mandate to one key, or let any key of the workspace present it.
Allowed actions
Create and prepare drafts, send for signature, remind signers, void, download, negotiate. Tick only what the agent needs.
Templates
Any template, or only the ones you approve. With an approved list, the agent cannot send a document of its own.
Sends per day
A daily cap, counted in your workspace’s time zone and reset at its midnight.
Recipients per envelope
The most people any one envelope from this agent may go to.
Recipient domains
Every recipient must have an address at one of these domains. Subdomains count.
Preview first
Every send, reminder or void through the MCP server must be previewed before it can be confirmed.
Expiry
An end date. After it, the mandate answers that it is inactive.
For negotiations
Which clauses the agent may negotiate, and how many turns it may take.How negotiation works

The agents console

Issue, Watch and Revoke From One Place

These are real screens from a sample workspace: its mandates, the run log of what their agents did, and the same activity on the Today page.

Mandates. Each row is one mandate: the agent, the key it is bound to, what it allows, today’s sends against the cap, its expiry, and the Usage and Revoke actions.

Swipe the screenshot sideways to see every column.

Allowed and refused

The Same Request, Inside and Outside the Mandate

The checks run on FlowVerify’s side, before anything is certified or emailed. An agent cannot widen its own mandate, and a refusal tells it which limit it met.

  • mandate_limit_exceeded

    The request is outside a limit. The reply names the limit and gives the value the agent needs: the cap and what is used, the allowed domains, the approved templates.

  • mandate_inactive

    The mandate is revoked, expired or bound to another key. The agent is told to stop and ask an admin.

  • mandate_required

    The workspace requires a mandate and this agent key did not present one.

What gets recorded

One Send, Written Down in Four Places

Everything an agent does under a mandate is recorded against the agent and, through the mandate, the person who issued it.

  1. Run Log

    Agents → Run log, one entry per request

    Sent “Vendor NDA — Nandi Freight” to Farah Qureshi

    16/11/2026 11:03:48 AM · FV-2611-102 · Claude, claude-ai/1.4 · Procurement agent

    Succeeded
  2. Audit Trail

    The document’s Activity tab and its audit trail PDF

    16 Nov, 11:03Sent by agent “Claude” under mandate “Procurement agent”, issued by Shalini Bose
  3. Event

    Your webhooks

    "type": "envelope.sent",
    "actor": {
      "type": "agent",
      "agent_label": "Claude",
      "mandate_id": "7f3a91c2-…",
      "issued_by": "Shalini Bose"
    }

Connected apps

Claude and ChatGPT Connect by Signing In

A connector cannot hold an API key, so the person adding it signs in to FlowVerify and decides on a consent screen. The screen names the app, lists each permission, and asks for live or test mode and for the mandate.

  • Approving needs the agents permission, and an identity check unless you have just signed in
  • Test mode connects an app without real emails or charges
  • Disconnect at any time under Agents, Connected apps
  • A connection ends when its approver leaves or loses the permission

By default, a connection ends after 30 days without use and after 180 days in any case. The app then asks you to sign in again.

Control

Stop an Agent, Check Its Usage, Make Mandates Compulsory

Revoke “Legal drafts”?

Agent “ChatGPT” stops on its next request.

Reason (optional)

The drafting project closed on 15 November.

CancelRevoke

Revoke With a Reason

Revoking stops the agent on its next request. What it already did stays recorded against the mandate, with your reason on record.

Procurement agent

Today, 16 Nov 2026

Sends

17 left today

3 of 20

Drafts
4
Reminders
2
Refusals
1

Latest runs

  • 11:03Sent “Vendor NDA — Nandi Freight” to Farah QureshiSucceeded
  • 02:11Refused: recipient domain gmail.com is outside the mandateFailed

See What Each Mandate Used

Sends against the cap, drafts, reminders and refusals, day by day, with the mandate’s latest runs a click away in the run log.

Require a mandate for agent keys

Live API keys marked as agent keys are refused unless they present a mandate. Other keys are unaffected.

An agent key without one getsmandate_required

Require a Mandate for Every Agent Key

Mark a key as an agent key when an agent holds it. With the workspace setting on, a live agent key is refused until it presents a mandate.

The admin. Issues the mandate, watches the usage, revokes

The agent. Prepares, sends, reminds and tracks, inside the mandate

The signer. Reads, verifies with OTP or Aadhaar eSign, and signs

An Agent Never Signs

No mandate can allow it. Signing is always done by the signer, in the signer app.

A first mandate

Start Careful, Then Loosen

A first mandate can be tight. Widen it once the run log shows the agent doing what you expect.

  1. Connect in Test Mode

    Use a test key, or choose Test mode when you approve a sign-in. Nothing is emailed, texted or charged, and the agent works only on test documents.

  2. Require a Preview

    Tick the preview setting. Every send, reminder or void through the MCP server is then previewed before it can be confirmed.

  3. Keep the Mandate Narrow

    One or two templates, your counterparties’ domains, a small daily cap and an expiry date.

  4. Read the Run Log, Then Loosen

    When the log shows the agent doing what you expect, issue a wider mandate, with a higher cap or without the preview, and revoke the first.

FAQ

Questions About Agents and Mandates

Something else on your mind? Book a demo and ask us.

Only the actions an admin ticked: create and prepare drafts, send for signature, remind signers, void, download and negotiate. Each can be narrowed further by approved templates, recipient domains, a recipient limit, a daily send cap and an expiry date. A mandate limits what an agent does. What it can read is decided by the permissions of its API key or connection.

The checks run on FlowVerify’s side, before anything is certified or emailed. A request outside the mandate is refused, and the refusal names the limit that was met. The agent cannot change its own mandate, and each refusal is counted on the mandate’s usage so you can see when an agent keeps reaching a limit.

Those of the person who issued the mandate, acting for your workspace. The certificate of completion states that the agent sent the document under that mandate, as an automated information system acting on the issuer’s instructions. Those are the facts section 11 of the Information Technology Act, 2000 looks to when it attributes an electronic record. Whether a sending binds your organisation depends on the authority behind the mandate.

No. An agent can prepare, send, remind and track. Signing is always done by the signer, in the signer app, with the verification the sender chose. No API call and no agent signs for a third party.

Revoke its mandate. The agent’s next request is refused, and what it already did stays recorded against the mandate, with your reason on record. For Claude or ChatGPT connected by sign-in, you can also disconnect the app under Agents, Connected apps.

No. Mandates are optional, and an API key without one keeps working as before. They are meant for the case where an agent holds the key instead of your own code. A workspace can also require a mandate for every live key marked as an agent key.

Book a demo

See It on Your
Own Agreements

Tell us which agent you want to connect. We issue a mandate with you, run a send inside it and a send outside it, and show you where each one is recorded.

What you get

  • A live walkthrough on one of your own agreements
  • A workspace set up for your team
  • Answers from the people who build it