Reports and agent workflows that run themselves

Each automation has one of three triggers. A schedule runs it at a set hour on a daily, weekday, weekly, monthly, or quarterly cadence. A data-change trigger runs it when a SQL query you write returns new rows. A manual trigger runs it only when someone starts it, from Basedash or through the REST API. Scheduled automations can also be started by hand at any time, which is the easiest way to test a change to the instructions.

You pick a frequency and an hour. Weekly automations also take a day of the week and monthly ones a day of the month, while quarterly runs land on January 1, April 1, July 1, and October 1. Basedash stores the hour in UTC and shows it in your local time, so a 9 AM report stays at 9 AM for you. If the previous run is still generating when the next one is due, Basedash skips the new one instead of stacking runs.

Basedash runs your query every 15 minutes, replacing {{ last_run_at }} with the time of the last run so you only see rows that are new since then. Each row that comes back starts its own run, with that row handed to the agent as context. A check starts at most five runs and skips any extra rows, so filter the query down to the rows that deserve attention. The query runs with the creator's data access and row-level security groups, and it can reference your models.

Each run is a full agent session with the instructions you wrote. It can query any connected source, build charts, and use the tools from MCP connectors that are set to Always allow, so it can add a note in HubSpot, file an issue in Linear, or send an email through Resend as part of the run. Tools that need approval stay out of unattended runs. Every run is saved in Basedash as a chat, so you can see each step it took and ask follow-up questions.

Every run is saved in the automation's history in Basedash. You can also send each result to one Slack channel, where its charts post as images, and to as many email recipients as you need. For a private Slack channel, invite the app with /invite @Basedash first. If you embed Basedash in your own product, the React SDK's BasedashAutomations component shows your automations and their runs there too.

An automation runs as the member who created it, with that member's data source access and row-level security groups, so a run can't read anything its creator can't. Only the creator can edit its instructions, which keeps someone else from changing what runs under their access. If the creator is deactivated, their data-change automations stop running. Customer data is never used to train AI models.

Yes. With an admin API key, the REST API can list, create, update, and archive automations, list their runs, and start a new run of a scheduled or manual automation with a single POST. Runs generate asynchronously: the POST returns the run right away, and you fetch it again to get its final answer as markdown along with its charts. Data-change automations can't be started this way, because their query decides when they run.

Each template in the library is a starting point for a common report, like a weekly MRR report or a closed-won deal alert. It includes the prompt, the trigger and cadence it's designed for, where it usually goes, and an example of the output. Copy the prompt into a new automation and adjust it to your tables and metrics. You can also skip templates entirely: describe what you want in chat and the agent sets up the automation for you.