Natural Language Input

Natural Language Input

The thing that makes jasem fast is that you never fill in fields — you write a sentence and jasem extracts the structure. This page documents exactly what it understands for dates, durations, and amounts.

There are two layers:

1. The AI parser (see AI Backends) turns your free text into rough fields — a title, a date phrase, a priority, tags, an amount, and so on.
2. A built-in, offline resolver then turns date phrases into real calendar dates and normalises durations and amounts. This layer is pure regex and always runs, so even with no model reachable, dates/durations/amounts are still parsed — you just lose the smarter title/tag splitting.

Dates

Dates are accepted when adding any entry and anywhere a deadline/date field is set. The resolver tries these forms, in priority order:

You write	Resolves to
2026-07-01 (any YYYY-MM-DD in the text)	that exact date
in 3 days, in 2 weeks, in 1 month	offset from today
next friday, next monday …	the next such weekday (always in the future)
tomorrow	today + 1
yesterday	today − 1
today, tonight, now	today
next week	today + 7
next month	one month ahead (day clamped to month length)
eow, end of week, this week	the coming Friday
friday, mon, this tuesday …	the next occurrence of that weekday — today counts if it matches
june 20, 20 june, june 20 2027	that calendar date; rolls to next year if the date has already passed and no year was given

Weekday and month names and common abbreviations are recognised (mon/monday, jun/june).

"No date" words

To mean no deadline, use any of: none, no deadline, no date, n/a, na, null, someday, - (or just leave it out). These also clear a deadline when editing: jasem todo set 3 deadline none.

How dates are displayed

Stored dates are absolute (2026-07-01) but shown relative so lists stay scannable:

• Deadlines: overdue, yesterday, today, tomorrow, in 4d, then in 2w (up to ~10 weeks), then in 3mo.
• Age (when a task was created): today, 6d, 3w, 5mo, 2y.

Durations (time tracking)

Used by jasem track. A duration is one or more number + unit pieces, summed:

Unit	Spellings
hours	h, hr, hrs, hour, hours
minutes	m, min, mins, minute, minutes

2h            -> 120 min
30 min        -> 30 min
1h 30min      -> 90 min
1h45min       -> 105 min   (units can butt straight against the next number)
90            -> 90 min     (a bare number is read as minutes)

A unit only counts when it's really a unit — the h in "5 hamburgers" is ignored. When a model is in play, phrases like half an hour (30) and an hour and a half (90) are understood too; the offline fallback handles the numeric forms above.

Durations are stored in a canonical 1h 30min form. If jasem can't find a duration at all, it saves your text as-is and warns that the entry won't count toward totals until you fix it with jasem track set <id> time ….

Amounts (spending)

Used by jasem acc. The first number found in the text is taken as the amount:

You write	Amount
50000	50,000
50k	50,000
1.5m	1,500,000
2b	2,000,000,000
1,200	1,200 (comma = thousands separator)
paid 50k for lunch	50,000

Magnitude suffixes are k (thousand), m (million), b (billion). Amounts are stored with thousands separators (50,000). If no number is found, the record is saved as-is and won't count toward totals until you fix it with jasem acc set <id> amount ….

jasem never assumes a currency — it sums plain numbers. Pick a unit and stay consistent.

Quoting and the parser

• Always quote a free-text description. It keeps multi-word text together and keeps shell characters (& ! * ( )) literal.
• The parser captures temporal words even after by/due ("submit report by friday") and pulls the duration/amount/tags out of the title, so the stored title stays clean.
• If your text starts with a command word (list, done, set, rm, …), see the force-add note on the Tasks, Time Tracking, and Spending pages.

See also

• AI Backends — choosing and configuring the model that does the first pass
• Jalali Calendar — typing and showing dates in the Persian calendar