A powerfully tiny agent for your personal analytics.
Point it at a folder of your files and ask in plain language. Every answer is a real calculation that shows its working.
How did my spending change this year?
fella
Your total spending is up 12% on last year, at $28,410
against $25,370. Most of the rise is groceries (+$1,900), travel (+$1,400) and utilities (+$310). Eating out fell $600.
- Listed the files in the folder
- Loaded
transactions.csv— 1,204 rows - Totalled spending by category for each year
select category, strftime('%Y', date) as yr, sum(amount) as total from transactions group by category, yr; - Compared this year against last, per category
$28,410 and $25,370 both appear
in a query result
category changes sum to the
reported total
llama3.1
A real computation over your own files — and it shows the working.
Numbers from queries
Every figure is real SQL over your files. The model never supplies one.
The working, always
One fold: files read, queries run, every figure re-checked.
It won’t guess
No answer in the data? fella says so — not an estimate in disguise.
Tabs and focus
Parallel conversations in tabs. Focus mode clears the chrome.
Read-only
fella reads the folder and nothing else. No writes, no prompts.
Yours to shape
Add themes, skills and connectors. Nothing is bundled.
Nothing leaves your computer.
The default model runs locally through Ollama, and every query runs on your machine. Only a hosted model you
pick sends anything out — its key kept in a 0600 file.