Automations
Gatlas alerts a Slack channel or an email address when a sync fails, posts a digest or an artifact on a schedule, runs a report for one company and keeps a dataset current in another tool.
Automations list
Open Automations, under Agents in the sidebar. The list holds every alert, scheduled Slack message, dataset copy, report and scheduled brief prompt you can see. Each row says what it does, which company it is for, when it runs, where it sends and when it last ran.
Anything that failed or stopped sits under Needs attention with the reason, and the rest under Running. Choose a row to change it on its own page. Sync schedules are not in the list: they are set on each connection, under Sources.
Create an automation
Choose what it should do
Choose New automation. Under What should it do?, pick one: Alert me when something happens, Post an artifact to Slack on a schedule, Send a Gatlas summary to Slack, Send a dataset to another tool, Have an agent write a report or Keep a brief current. You see the ones your role can set up.
Choose who it is for
Under Who is it for?, choose Whole workspace or One company and the company. When only one answer is possible, Gatlas skips this step.
Fill in its settings
The last step is the same form the automation's own page uses, and its button says what happens: Turn on alerts, Schedule post, Create dataset copy or Run report. The new row then appears in the list. Keep a brief current is the exception: Continue in Brief prompts opens a new brief prompt for the company, set to run weekly.
A Slack automation needs the company's Slack channel. If it has none yet, the last step offers Connect Slack and Create install link, and Slack brings you back to the same step once the channel is connected.
Connect Slack
You can connect Slack for the whole workspace and for each company. The owner connects the workspace and each top-level company; admins and lead admins can also connect a company nested under another. Other roles do not see these settings.
A Slack connection and a report cover a company and everything under it, so setting one up takes access to all of that. Someone whose access is Only specific companies cannot manage Slack or reports, whatever their role.
Choose where alerts should go
Open Automations, under Agents in the sidebar. Under Slack channels, the table shows your workspace and your business, or each company, with its Slack channel and a status of Not connected, Connected or Paused. Choose one.
Connect Slack yourself, or send an install link
Connect Slack takes you to Slack to approve Gatlas and pick a channel. If someone else administers your Slack, choose how long the link should last under Install link expires in (1, 3 or 7 days), then Create install link. The link is copied for you to send; they need no Gatlas account.
Send a test
Once connected, Send test posts a message to the channel. If Gatlas cannot post there, invite @Gatlas to the channel. Change channel, Pause and Disconnect sit beside it.
Alerts
Choose New automation, then Alert me when something happens. For the whole workspace, an owner chooses Slack or Email under Where should alerts go?, and an admin goes straight to Email. That adds a recipient for sync failures, connections not syncing and connections that need reconnecting. Change it later under Email alerts at the foot of the page.
In Slack, tick the events Gatlas should post about and choose Turn on alerts. When you connect a channel, every event except Unusual syncs starts ticked, and the alerts row in the list opens them again. The events:
- Sync failures
- Connections not syncing
- Connection needs reconnecting
- Unusual syncs
- New connections
- New members
- New agent keys
- Sync recovered
- A brief stopped updating
- Agent key first used
- Agent key revoked or changed
- Agent feedback
- Connection removed
- Member removed
Unusual syncs posts when:
- a sync brings in more than three times a connection's recent average, less than a fifth of it, or nothing at all
- a tool's own row count for a table and Gatlas's disagree several syncs in a row
- a table has had no new activity for more than seven days, while the connection's other tables kept syncing
For a quiet table, it posts when the tool gives a reason you can fix, or shows activity Gatlas has not read. It also posts when the tool gives no reason for a table that usually gets new activity.
A reason nobody needs to act on, such as a paused campaign in Google Ads, sends nothing. A quiet table posts once until it gets new activity again.
Keep Sync failures, Connections not syncing and Connection needs reconnecting on at least. Without these three, a broken connection can go unseen until someone asks a question it cannot answer.
Scheduled Slack messages
Choose New automation, then Post an artifact to Slack on a schedule or Send a Gatlas summary to Slack. A summary is one of these:
- Weekly digest: connections and syncs, what agents asked, briefs updated, reports run and automations paused in the last seven days, with a sample of approved datasets.
- Usage summary: what agents asked Gatlas in the last 24 hours, by kind of request. It is skipped on a day with no agent requests.
- Source health: which connections are failing or have had no sync in 2 days.
An artifact post sends one of your artifacts, with a short intro and an Open artifact button.
Then set the Cadence (daily, twice daily, weekly or monthly), the Send time, the Timezone, and the Channel if it should go somewhere other than the connected one. Preview shows the message before you choose Schedule post or Schedule summary. A row in the list opens the channel's scheduled messages, where you edit, switch off or add more. A channel connection holds up to 20 scheduled messages.
Every new channel comes with two: a usage summary every day at 09:00 UTC, switched on, and a weekly digest on Mondays at 09:00 UTC, switched off. Edit or switch off either one.
Tip
{artifact_name}, {artifact_description}, {scope_name} and {date}. Monthly board packs work well this way: build the artifact once, then schedule it for the first of the month.Reports
A report puts one question to an agent that reads a company and everything below it: its synced tables and files, approved datasets, briefs and AI context. It cannot reach anything outside Gatlas or change anything, and every answer cites the data it used. It runs each time someone runs it; nothing repeats it on a schedule.
Choose the company
Choose New automation, then Have an agent write a report, and pick the company. A report is for one company, never the workspace as a whole, and needs no Slack channel.
Ask the question
Fill in Reporting question, for example "How did gross margin develop this quarter, and what drove it?", and check Reporting timezone. Additional context is a note for the people reading the report later; the agent never sees it. Choose Run report.
Read the result
The run is queued, and Open report in the confirmation opens it. The report's row in the list opens Reports for the company, with its Recent runs. Open a run to read the report and the data each section used, choose Run again for a fresh answer, or Revoke sign-off to stop it, which also stops a run that has not finished.
Note
Dataset copies
A dataset copy pushes a dataset into another tool, its destination, and keeps it current there: a Google Sheet, a Slack channel or a webhook. On the Automations page, choose New automation, then Send a dataset to another tool. Any dataset offers every dataset you can see, and One company's datasets only those tagged to that company.
Give it a schedule, hourly, daily or weekly, and Gatlas runs it that often. Its page says when it next runs. After a failed run Gatlas waits longer before trying again, and five failed runs in a row stop the schedule until a run succeeds or someone changes the schedule. An admin restarts a stopped schedule with Run now.
Verify a dataset copy webhook
A dataset copy can send a dataset to a webhook. When its destination has a webhook secret, every delivery carries two headers. Gatlas-Timestamp is the time of the send in Unix seconds. Gatlas-Signature is v1= followed by the hex HMAC-SHA256, keyed by the secret, of the timestamp, a period and the exact request body. A delivery without a secret carries neither header, and a Slack destination is never signed.
To check a delivery, compute the same signature over the raw body before parsing it, compare the two in constant time, and reject a timestamp more than 5 minutes from your clock, so an old delivery cannot be replayed. In Python:
import hashlib, hmac, time
def verify(secret: str, body: bytes, timestamp: str, signature: str) -> bool:
try:
if abs(time.time() - int(timestamp)) > 300:
return False
except ValueError:
return False
expected = hmac.new(secret.encode(), timestamp.encode() + b"." + body, hashlib.sha256)
return hmac.compare_digest(("v1=" + expected.hexdigest()).encode(), signature.encode())