How it works

Connect and configure your agent

Set up logging, test pre-action protection and choose how your data is handled.

The pause button can only stop actions that ask Groundstop first and obey the response. Logging after an action cannot prevent it. Check before emails, cancellations, refunds, deletions and payments—not necessarily before every step. An action already running cannot be stopped or undone.

1. 01

   Create your account

   Sign up with email or Google and accept the Terms of Service and Data Processing Agreement. You do not need an AI key to record activity, receive rule-based alerts or use pre-action checks.

2. 02

   Connect your first agent

   When you have no agents, the dashboard starts with “Connect your first agent in 2 minutes”. Enter a recognisable name, pick n8n, Make or Zapier and select “Create agent”. The agent's page opens with your platform's ready-made template selected. Copy it, connect it to your automation, then return to the dashboard and press “Test” to see an entry in the flight log. The flight log's legend explains its dots: green means OK, red means failed, blocked or a broken rule, and every red entry states its reason in the meta line (for example FAILED, CHECK BLOCKED or OVER LIMIT). Once an agent exists, the dashboard shows stats, alerts, the flight log and your agents. Add further agents in the Agents section. Each agent's links contain a secret token: anyone with the token can send requests for that agent. Keep it out of public screenshots and shared workflows.

3. 03

   Explore with sample data if you prefer

   Select “Load sample data” on the empty dashboard to create “Support agent”, with 25 entries from the last 24 hours and six alerts. The sample includes customer replies, CRM updates, refunds, a cancelled order, an email burst, failures and a blocked check while paused. Sample badges identify these entries throughout the dashboard and agent history, and the agent's charts show the sample activity. Samples never send external notifications or use AI. Select “Remove sample data” in the Agents section to remove the sample agent and all its entries and alerts without affecting your real agents.

4. 04

   Choose logging or pre-action protection

   Log: send the actual result to Groundstop after an action. Protect: place “Ask Groundstop first” immediately before a sensitive action, add a filter that continues only when allow is true, perform the action, then “Log to Groundstop”. Stop when permission is denied, the check fails or there is no valid response. Use the platform template as a starting point and connect it to your own action.

5. 05

   Send minimal, accurate event data

   Describe the action, identify the target with a reference such as order #4790, and include the amount when relevant. When logging the result, send status “ok” or “error” and only the details needed to understand it. Avoid customer records and secrets. Personal-data masking is on by default; email bodies, HTML, attachments and base64 blobs are removed, and long text is truncated. Automatic masking is not a guarantee that every personal detail will be detected.

6. 06

   Send monetary values in US dollars

   Spending, log amounts, alert amounts, per-action limits and monthly budgets are displayed in US dollars ($). The “AI spend · month” stat shows what your agents cost in AI calls—not money the agent moves, such as refunds. Send cost_usd for an action's cost, or cost as an alternative. The older cost_eur field remains accepted so existing connections continue working. Send amount for the transaction value used by amount-limit checks; if no cost field is supplied, amount is also used as the recorded cost. If you send several cost fields, priority is cost_usd, then cost, then cost_eur. Values are not converted between currencies: 12.50 is displayed as $12.50, including values sent through cost_eur. Convert non-dollar values in your workflow before sending them.

7. 07

   Set the rules for this agent

   Configure sensitive words, the amount limit, email bursts and repeated-error thresholds on the agent's page. Set a monthly budget and expected activity interval if relevant. Alerts flag recorded activity; they do not undo it. Pre-action checks deny permission when the agent is paused or the amount exceeds its limit. Enable “Block risky actions” if matching sensitive words should also deny permission; it is off by default. An allowed check is recorded as a green “CHECK ALLOWED” entry without an alert—only denied checks and rule-breaking actions create alerts, so you are not notified twice about the same action.

8. 08

   Test logging and protection separately

   Use “Test” to check that an entry appears in the flight log, then run your own workflow. This dashboard test does not prove your automation is connected or protected. For protection, verify both an allowed action and a denied one: pause the agent and confirm that the workflow stops before acting, then resume it. Also confirm it stops when the check cannot be reached. A Protected badge means recent check or status contact within 24 hours—not proof that every action is protected. Greyed-out buttons indicate an unavailable action; for example, Create agent needs a name and Test is unavailable while the agent is paused.

9. 09

   Add AI only if you want it

   Without a valid AI key, the dashboard shows one compact “Unlock AI features” card at the bottom instead of the question and briefing panels. Follow “Add key in Settings”, choose OpenAI, Anthropic or Google Gemini, enter your own API key and save it. A model is optional. Once the key is valid, Ask Groundstop and AI Daily Briefing appear. AI-written summaries, questions, briefings and AI incident analysis use your provider account, which bills you directly. Without a key, Groundstop uses deterministic summaries and its logging, alerts and pre-action checks still work. AI summaries are processed while the dashboard is open; if the key fails, logging continues with non-AI summaries.

10. 10

    Choose retention and manage your data

    In Settings, keep “Redact personal data” enabled unless you have a specific reason to turn it off. Choose “Keep logs for” 7, 30 or 90 days; older logs are removed by a daily cleanup, not at the exact expiry minute. Changes do not rewrite previously stored data. Use “Export my data” for JSON or CSV, “Delete all logs” on an agent's page to clear its history, or “Delete my account” to permanently erase your account and its stored data.