Instrument 01 · Hypothesis printer
Ask, don’t build.
Describe the board you want. The agent writes the SQL and checks every query.
Version 0.1.15 for Mac
Follow your weirdlittle hypothesis.
Probably nothing. Let’s plot it. Connect your tools, ask in plain English, and Etlo draws the chart before the hunch wears off.
Download for MacGitHub
Fig. 1 · Live specimen
This is the real app. Poke it.
Sample data, real engine. Filter, sort, sync, or ask the agent for a new chart.
What do you want to know?
Synced
§1 Apparatus
You already own most of the equipment.
Pick from the catalog, or let the agent connect any API.
Specimen 0047, between meals
- Linear
- Stripe
- HubSpot
- Notion
- Shopify
- Zendesk
- GitHub
- Jira
- MailChimp
- Airtable
- Google Sheets
- Intercom
- QuickBooks
- Xero
- Asana
- Trello
- ClickUp
- GitLab
- PostHog
- Mixpanel
- Google Analytics
- Sentry
- PayPal
- Calendly
- Webflow
- Gumroad
- Paddle
- Lemon Squeezy
- Plausible Analytics
- Figma
- Gmail
- Zoom
§2 Method
The scientific method, abridged.
Three steps. One afternoon. The third one repeats itself so you don’t have to.
- Step 01
Hypothesize
Say what you suspect, in a sentence. Weird is fine. Weird is encouraged.
- Step 02
Plot
The agent connects the data, writes the SQL and draws the chart.
- Step 03
Rerun
Boards repeat the experiment on a schedule, without calling the model.
§3 Instruments
A whole research department.
One app.
Connectors, models and charts, built by the agent and kept fresh.
Instrument 02 · Sample fridge
Only what changed refreshes.
Charts are cached against their data, so boards open with no API calls.
Instrument 03 · Lab notebook
Every change is a file.
Plain JSON and SQL with history. Undo any agent edit in a click.
Instrument 04 · SQL scope
Every number has a source.
Open the SQL behind any chart and read exactly how it was counted.
Abstract
§4 Field work
Any API.
Field tested for you.
Paste the docs. The agent writes the connector and tests it.
§5 Facilities
Start on the bench.
Grow into a whole lab.
One engine runs four ways, with the same project files wherever it lives.
§6 Results
Peer reviewable.
We tried to break it so you don’t have to. Pull the plug halfway through a sync and the next run resumes from its last checkpoint, with no duplicates and no gaps.
- Duplicate rows after a sync is stopped halfway.
- 0
- Gaps after it resumes from the last checkpoint.
- 0
- Model calls during scheduled refreshes.
- 0
- API calls to open a board.
- 0
§7 Peer review
Questions from the audience.
Short answers, peer reviewed. The longer ones live in the app.
Where does my data live?
On your Mac, in an encrypted DuckDB file. Credentials sit in an encrypted file whose key is kept in the macOS keychain. We run no servers that hold your data.
What does the AI agent see?
Table schemas and the results of the queries it runs to check its work. Those go to the model provider you pick. Scheduled syncs and refreshes never touch the model.
Do I need an API key?
Yes, an Anthropic or OpenAI key for the agent. It is stored in your keychain. Syncs and dashboards keep running without one.
Which tools can I connect?
Pick a connector from the catalog, or paste a link to any REST API’s docs and the agent writes one, then tests it on real data before it syncs.
Can I edit what the agent built?
Yes. Every source, model and dashboard is a JSON or SQL file with git history. Each agent request is one entry you can undo in a click.
What happens when my laptop sleeps?
Schedules run while your Mac is awake, and each board shows when it last synced. A sync stopped halfway resumes from its last checkpoint.
How much does it cost?
Etlo is free, with no account. You only pay your model provider for the agent’s requests.
Windows or Intel Macs?
Not yet. Today’s build needs macOS 13 or later on Apple Silicon.
§8 Conclusion
Got a hunch?
Plot it tonight.
Free. No account. Starts on sample data, so you can experiment first.
Download for MacGitHub