Lookbacks & AI Context

Lookbacks

Every day Loopr generates a daily lookback — a short markdown reflection on what happened. It lands in your loop like any other item; open it to read.

The daily covers:

Summary — 2–4 sentences on what the day actually looked like
Highlights — the notable concrete items you completed
What slipped — things that sat untouched, with drift signals (“14d old,” “bumped 3× without being completed,” “previously snoozed”)
On the horizon — items with approaching due dates
Delegated & outstanding — what you’re waiting on from others
Mindfulness & routines — what you did vs. skipped
Quiet observations — at most 3 gentle patterns the AI noticed

Weekly, monthly, and quarterly lookbacks are generated from the level below them — the weekly reads the seven dailies, the monthly reads the weeklies, the quarterly reads the months. This keeps inputs bounded and makes higher-level lookbacks genuinely lookback, not re-read everything.

Lookbacks are reflective, not prescriptive. They won’t tell you what to do next. They describe. You decide.

Context suggestions

At the top of each lookback, you may see a small callout: “The AI suggests adding this to your context.” That’s the model noticing — while writing — that it lacked a specific piece of background that would’ve let it make sharper observations. One sentence, actionable.

Two buttons:

Edit and add to context — opens a textarea pre-filled with the suggestion, lets you modify it, then appends it to your AI Context
Dismiss — drops the suggestion

The model is instructed not to produce generic filler (“tell me more about yourself”) — only specific gaps it actually noticed.

AI Context

Open the Profile modal and you’ll find an AI Context textarea. Anything you put here gets threaded into the system prompt of every AI action in Loopr — summaries, lookbacks, urgency ramp, triage — as a consistent preamble about you.

Good things to put in:

Your role, what you do, who you do it for
Current priorities (projects, initiatives, the season’s focus)
People who matter (names and relationships)
Working constraints (“I work mornings only”, “I’m in EST”, “newsletters are always low priority for me”)
Strong preferences (“I don’t want advice, I want observations”)

The context is stored as markdown, so you can structure it however makes sense. The AI is told to “apply it silently” — it won’t parrot your context back at you or say “according to your context…”; it’ll just use it.

Because the preamble is stable across requests, it hits prompt caching on the model side. Editing it doesn’t cost you anything in terms of output quality; large context blocks don’t make later calls slower.

Scheduling

Lookbacks are generated on a schedule:

Daily lookback at end-of-day (respects your configured day-end time)
Weekly on Sunday nights
Monthly on the last of the month
Quarterly at the end of each quarter

If there’s nothing to summarize (e.g. you hadn’t used Loopr that day), no lookback is generated.