Grant Ledger

Simulated: the agents, the services, their replies, and the clock are fake. No real agent or service takes part.

Try this

  1. Press Ask for access. When an agent asks in the chat, press a Grant button.
  2. Watch the receipt stream and the Receipt JSON panel. Set the clock to 60x and wait, or press Revoke.
  3. Press Agents try again. Read the Denied receipt.

Clock:

Chat with your agents

Grants

    Receipt JSON (newest receipt)

    No receipt yet.

    Receipt stream

      How it works

      Who checks a grant, and when Four actors: the agent, you, the identity provider, and the resource server. Step 1: the agent asks you. Step 2: you grant, and the identity provider stores the end time. Step 3: the agent uses the grant at the resource server. Step 4: the resource server checks the grant and writes a receipt. Step 5: the identity provider ends the grant, and the next use gets a Denied receipt. Agent Resourceserver Identityprovider You 1 2 3 4 5 Numbers match the steps. The check at step 4 happens at every use.

      Pain point

      People cannot easily see, limit, or end what an agent can do for them.

      Problem statement

      An agent asks for access, and the person has no clear view of what it holds. Grants do not end by themselves.

      What sparked it

      Mozilla's W3C WebExtensions group notes (issue 1041) that no browser-level primitive exists for a user to see, scope, or withdraw an agent's access.

      Why this approach

      Each grant has a task, a scope, and an end time. A grant ends by itself, and every use writes a receipt.

      Hypothesis tested

      We believed that a grant with a task, a scope, and an end time gives a person control. If it holds, people can tell what an agent may do and when that access ends.

      How to try it

      1. Open the Prototype tab and press Ask for access.
      2. Press a Grant button, or press Deny.
      3. Set the clock speed to 60x and press Revoke, or wait for the grant to end.
      4. Press Agents try again and read the Denied receipt.

      What this does not do

      Notes

      Traces

      Prompts used to make this prototype
      StepPromptModelToolDate

      Cost

      Cost to make this prototype
      StepModelInput tokensOutput tokensCost (USD)
      Total

      Measured: the author runs used 203,100 tokens in total. Estimated: the split into input and output tokens (90 percent input, 10 percent output) and the dollar cost, at $2 input and $10 output per million tokens for Sonnet 5.5 (about $0.57). The run reported one total only. The lead session's planning, research, and publishing are not included. The prompts in the Traces section are listed in the order they were given.