Daily summaries

Overview

Daily summaries turn a day’s Overshow captures into a readable digest: what stood out, what was decided, what you might want to remember, and where on-screen time concentrated across applications. Generation runs on device over hourly aggregates of capture data, then persists the result encrypted at rest. You can open any past date that has capture data and generate on demand from the desktop app, with a progress indicator while work completes.

What a daily summary contains

The digest is evidence-led: it reflects what appeared in your captures, not a separate time-tracking product or cloud summarisation service for the core flow.

Summary sections

Section	Typical content
Key facts	Notable statements, figures, and concrete details that showed up in OCR or transcribed audio during the day.
Decisions	Commitments, choices, and conclusions that can be inferred from captured discussion and on-screen work.
Learnings	Insights, corrections, and “things worth remembering” distilled from the day’s indexed material.
App usage breakdown	Which applications appeared in screen capture and how capture time allocated across them for that calendar day.

Relationship to hourly capture

The summary pipeline consumes work driven from hourly capture aggregation. the same underlying rhythm that organises much of the desktop index. That is why a day with sparse or uneven capture may yield a shorter digest: there is simply less signal per hour to map and reduce.

How generation works

How the day is processed

Stage	Role
Hourly extraction	Each hour contributes candidate facts, decisions, and learnings from that window’s captures.
Merge	The pipeline merges hourly contributions into a single daily narrative with consistent tone and without endless repetition.
App usage	Application names and durations are derived from which apps were visible in screen captures across the day, then presented as a breakdown alongside the narrative.

Worker and backend

The summary pipeline processes pending work items locally and writes results back for the UI to display.

Backend	Role
LLM	Produces richer phrasing and connection between ideas when a language model is available in your configuration.
Baseline fallback	Ensures you still receive a structured digest when the LLM path is unavailable or errors; expect simpler linkage between bullets.

Enabling daily summaries

Daily summary generation is guarded by a feature flag, which is off by default. Enable it in product settings before expecting automatic or background generation to run.

Mode	What to expect
Flag off	No background generation for daily rows; the UI may still show historical rows if they were created earlier.
Flag on	Worker can pick up pending daily summary jobs according to schedule and datastore state.

First-time setup checklist

1. Turn daily summaries on where your build exposes feature flags.
2. Confirm capture is running for the days you care about.
3. Open /summary, pick a date, and use generate on demand to validate output before relying on background completion.
4. Review ignored applications so important tools are not excluded from app usage and OCR signal.

Generating on demand

Even when background scheduling exists, you remain in control: the desktop app lets you request generation for a chosen date. A progress indicator reflects worker activity so long-running days do not feel stuck.

Situation	Suggestion
You were away	Pick the date with real capture activity and generate; past dates remain available whenever data exists.
A day looks empty	Confirm capture was not paused all day and that ignored apps did not hide your main work surface.
You want a second opinion	Regenerate after changing capture or ignore settings so the next pass sees different evidence.

Desktop summaries experience

The /summary route is the primary place to read and trigger daily digests.

Element	Purpose
Date picker	Jump to any day with capture data; compare adjacent days when reconstructing a multi-day decision.
Knowledge cards	Present facts, decisions, and learnings in scannable blocks aligned to the table above.
App usage breakdown	Shows which applications dominated visible screen time in captures for that date.
Progress UI	Surfaces generation state while the worker runs so you can wait or navigate away intentionally.

What app usage measures

Measures	Does not measure
Visible application identity in frames	Keystroke or input volume
Temporal spread of captures per app	Silent background CPU
Honest signal when capture runs	Time when capture paused or excluded

Encryption and storage

Completed daily summaries are stored encrypted at rest alongside other sensitive narrative content in your local datastore. This matches the product’s privacy-first posture: the digest is yours, readable on device, and not dependent on a third-party summarisation SaaS for the default path.

When summaries can be thin

Some days legitimately produce short or generic output. Common causes:

Cause	Effect
Low capture volume	Few hourly buckets contain OCR or audio text to map.
Paused capture	Large gaps remove context the reducer could connect.
Ignored or excluded apps	Work happens in tools that never enter the index.
Headless or minimal UI	Little on-screen text for OCR to anchor facts.
LLM unavailable	Baseline fallback may produce flatter prose with less cross-hour linking.

Diagnosing a thin summary

Compare capture settings, pause history, and ignore lists for the date in question. If app usage shows only one or two applications yet you remember more context, the missing apps were likely not captured or not visible in frame samples. Regenerate after a representative day to confirm improvement.

Tips for richer summaries

• Run capture consistently during working hours you care to remember.
• Keep ignored windows limited to genuinely sensitive or noisy surfaces.
• Prefer visible work in standard desktop apps when you need strong OCR anchors (browser tabs, documents, tickets).
• Enable daily summaries when you want background completion; use on-demand generation to backfill specific dates.
• Compare adjacent days when rebuilding a narrative across a week. daily summaries are per calendar date, not arbitrary ranges.

Viewing past dates and comparing days

The date picker is not limited to “today”. Any calendar day for which capture data exists in your local database can be selected, and you can generate or regenerate a summary for that day.

Workflow	Why it helps
Catch up after leave	Reconstruct decisions and facts without manually scrolling search results.
Audit a specific day	Pair the digest with Search filters for time range when something is disputed.
Week-over-week review	Open adjacent dates in sequence; app usage patterns often explain shifts in tone or focus.

If a date is missing from the picker or cannot generate, there may be no capture rows for that day, or storage was reset on this device. summaries cannot invent content that was never indexed.

LLM versus baseline (practical differences)

Aspect	LLM	Baseline fallback
Wording	More natural connective language between hours	More list-like, fewer implicit links
Risk of overreach	Guardrails still apply; content should tie to captures	Tends to stay closer to literal fragments
Availability	Depends on platform, model, and settings	Intended to always be available as a backstop

You do not need to pick manually: the pipeline selects the LLM path when appropriate and falls back otherwise.

See also

• Meetings for call-level narratives and templates.
• Search to drill from a bullet to evidence.
• Screen capture for how frames enter the index that feeds summaries.