Check the work
Check the work is a second check before anything goes out. Put it right after an xMap LLM or an AI agent (or any block): it checks each item against your rules in plain words and ready-made checks, and what fails is sent back to be redone with what was wrong, removed, or kept and marked. Each item gets check_passed (Yes or No) and check_issues.
In the editor: A second check before the work goes out. It checks each item against your rules in plain words (every lead has a phone, never invent numbers, formal tone, under 120 words) and ready-made checks. What fails can go back to the xMap LLM or AI agent before it to be redone with what was wrong, be removed, or stay marked. Each item gets check_passed (Yes or No) and check_issues.
Settings
Rules
Text, several linesOne rule a line, in your own words: every lead has a phone, never invent numbers that are not in the data, formal tone, no emojis, the email is under 120 words. What code can check (a phone, a website, an email, numbers, personal data, word limits, emojis) is checked by code, for free; the other rules are judged by xMap AI.
Rules code cannot check are judged by xMap AI.
Ready-made checks
Pick one or moreChecks you can tick instead of writing them: every item has a phone, a website or an email; no numbers that are not in the item’s own data; no personal data (email addresses, phone numbers, ID-like numbers) in the text; written in one language.
- Every item has a phone
- Every item has a website
- Every item has an email
- No numbers that are not in its data
- No personal data in the text
- Written in one language
Language
Pick one- English
- Arabic
- French
- Spanish
- Portuguese
- German
- Italian
- Dutch
- Turkish
- Japanese
- Chinese
- Korean
- Hindi
- Indonesian
- Russian
What to check
Pick oneAutomatic checks the fields the xMap LLM or AI agent before it wrote. Fields I choose checks the fields you name. The run’s report checks the one report an xMap LLM wrote about the whole list.
The fields the xMap LLM or AI agent before it wrote; after any other block, the item’s longer text fields.
- What xMap AI wrote (automatic)
- Fields I choose
- The run’s report
Fields to check
TextNeeded before the flow can run.
Report
TextThe name of the report, as in {{report}}.
Needed before the flow can run.
If it fails
Pick oneSend it back: only the items that fail go back to the xMap LLM or AI agent right before this block, with what was wrong on each, and are checked again. Keep only the ones that pass, or keep everything with check_passed saying what failed.
Sent back to the xMap LLM or AI agent right before this block, with what was wrong, and checked again. Its own credits are paid for each item redone.
- Send it back to be redone
- Keep only the ones that pass
- Keep all, marked
Send it back at most
Pick oneHow many times an item may go back. Each time is paid at the price of the AI block that redoes it.
- Once
- Twice
- 3 times
Still failing after that
Pick oneWhat happens to items that still fail after the last try.
- Remove them
- Keep them, marked
What comes out
Each item that leaves the block carries these fields, to use in the blocks after it as {{field}}.
Each item
Every item that reaches it (the ones removed are left out).
check_passed: Yes when it met every rule, No when it did not.check_issues: What was wrong, in plain words.check_redone: How many times it was sent back, when it was.
How to use it
Pitches checked before they go out
Every pitch the AI writes is checked before it reaches the team; the ones that fail are rewritten with what was wrong.
Only leads you can call
Leads without a phone number never reach the CRM.
Replies with no personal data
Each answer is checked before it is sent; one that fails is written again.
Listing summaries that stick to the facts
A summary that gives a price or a size the listing does not have is sent back.
A weekly report checked
The brief goes out only when it keeps to the figures your AI employees found.
Answers in one language
Every answer is in the language the service promises.
Good to know
- Checks that code can make cost nothing. Rules only words can say are judged by xMap AI, a credit for each item it reads.
- Send it back works with an xMap LLM or an AI agent right before the block (a Filter or a Sort between them is fine). An agent’s answer that goes back is not remembered twice. When the redone answer is kept, what its tools queued (an email, a message) is sent in place of the first answer’s, never both, and never more than the first answer’s sends were held for. An answer the check removes takes with it what the agent’s tools were to send: it is not sent.
- An item that still fails after the last try is removed or kept marked, as you choose; the step says how many were redone and what became of them.
- Test runs check and redo too, and send nothing, as always.
- In the editor’s preview the code’s checks run, and the rules for xMap AI pass until a run on the server judges them.
Works well with
- xMap LLM: Tell the AI what to do in plain words.
- AI agent: An AI agent: for each incoming message it thinks, uses the blocks you link to its Tools port (data, xMap AI, lookups, your tables, other flows, apps), remembers the conversation and writes an answer in {{reply}}.
- Email: Sends one email per run with all the items, or one email per item.
- Reply: Answers whoever started the flow, on the same channel: the app or system that called its Webhook / API gets the answer back, and a message from Slack, WhatsApp, Telegram, Teams, SMS, Outlook or Discord gets a reply in the same chat, through the account of the block that received it.
Help articles
- Check the work before it goes out: A second check of what xMap AI wrote: your rules, ready-made checks, and redoing what fails.
Read more
- How much autonomy to give an AI agent, decided one action at a time: How much autonomy to give an AI agent: split its work into looking, deciding and acting, set limits on each, and widen them only as its track record grows.
- AI employee approval before sending: who signs off, how long it waits, and when to stop asking: A sign-off step is only useful if someone answers it. How to pick the approver and the wait, put an automatic check first, approve from your phone, and loosen it later.
- Measuring the work of AI employees: the right number for each kind of job: A lead finder, a watcher and a chat agent need different yardsticks. Which numbers to read for each, what hours saved really means, and a monthly routine that takes minutes.
Try it
Build a flow with Check the work on the home page, no account needed, or book a call. Try it live