Saemi is an AI-native data workspace. Connect your warehouse, ask a question in plain English, and the answer becomes a chart, a wiki, and a reusable skill — all from one chat surface.
This guide walks through the loop once so you know what to expect.
1. Sign up
Open saemi.at and click Start free. Saemi uses Google SSO — no email/password to remember. You'll land on a one-screen setup that asks for your workspace name and team size.
2. Connect a data source
After setup, Saemi auto-attaches a demo Postgres database so you can try the loop without any of your own data. To connect your real warehouse, open the connector at /accounts/connector/ and pick from:
- Databricks — workspace URL + personal access token
- Snowflake — account, warehouse, username, password
- BigQuery — service-account JSON
- PostgreSQL / MySQL — standard host / port / user / db credentials
- Amazon Athena — region + S3 staging bucket + AWS credentials
Saemi tests the connection on save and runs a brief schema scan to power table autocomplete in chat.
3. Ask your first question
Click New chat and type a real question — Saemi expects natural language, not SQL. Examples:
- "What's our W2 retention by acquisition channel last month?"
- "How many active accounts opened a chat this week?"
- "Show MRR by plan, by month, for the last year"
Saemi writes the SQL, runs it against your warehouse, returns a table, and offers to render it as a chart.
4. Save the answer as knowledge
Below every answer is a Trust bar. Click Save as wiki to turn that answer into a permanent wiki page — the next time someone asks the same question, Saemi cites your wiki instead of re-running the query.
You can also click Add fact to teach Saemi a definition (e.g., "active account = made ≥3 chats in the last 30 days"). Saved facts are checked into every future answer that touches the same metric.
5. (Optional) Turn it into a skill
If the same question comes back every week — daily standup metrics, weekly investor update — open the Skills tab and build a skill. A skill bundles a prompt, the data sources it needs, and an output template (wiki, dashboard, Slack message). One click reruns the whole flow.