AI Summary
AI Summary is no longer a single global prompt with one generic generation step.
It now starts from insight cards: choose an analysis angle first, then set range, visibility, and prompt template before running the report.
1. What the new AI Summary looks like
The current AI Summary home shows 9 template entries:
- 8 built-in insight cards
- 1 custom template card
Tapping a card opens the analysis settings sheet, where you decide whether to start analysis.
2. The current 9 summary templates
| Template | What it helps you uncover |
|---|---|
| Today Clues | The most important recent events, issues, and unresolved signals |
| Emotion Map | Emotional movement and factors that are repeatedly affecting your state |
| Theme Resonance | Repeated themes that show what keeps drawing your attention |
| Thought Trace | Ongoing ideas, open questions, and hidden assumptions |
| Blind Spot Discovery | Signals, details, or patterns that are easy to miss |
| Relationship View | How recent notes look from a people-and-relationships angle |
| Action Extraction | Next actions that can make reflection more actionable |
| Long-Term Trajectory | Trends, loops, and stage changes across a longer time span |
| Custom Template | Your own title, description, and prompt template for personalized analysis |
3. What must be ready before analysis
Required: a working Chat model
Without a working Chat model, AI Summary cannot start.
Recommended: an Embedding model
An Embedding model is optional, but recommended:
- With embedding: the app can do vector retrieval and evidence recall before generation.
- Without embedding: the app still works, but falls back to direct reading mode and analysis accuracy may be lower.
Optional: My Profile
My Profile acts as background context for analysis. It is optional and does not block AI Summary.
Extra requirement for the custom card
If you use the custom template card, configure all of these first:
- Title
- Description
- Prompt template
4. What you can configure for each run
Inside the settings sheet of each insight card, you can currently adjust:
Time range
- Last 3 days
- Last 7 days
- Last 30 days
- Custom range
Visibility scope
You can independently include:
- Public (
PUBLIC) - Private (
PRIVATE) - Protected (
PROTECTED)
This is more granular than the previous private-notes-only switch.
Prompt template
- Each built-in insight card has its own default prompt template.
- You can edit the template for a specific card instead of changing one global prompt.
- The custom template card is fully defined by you.
5. How notes are selected for analysis
AI Summary reads memos from the local database, not directly from the server UI.
Before analysis, memos are filtered by:
- selected time range;
- selected visibility scope;
- non-empty content.
After that, the app runs in one of two modes:
When an Embedding model is available
- the app prepares vector indexes;
- retrieves evidence snippets around the current insight card;
- builds an evidence pack for the chat model to analyze.
When no Embedding model is available
- the app falls back to direct reading mode;
- the summary can still run;
- retrieval quality is weaker, so analysis may be less accurate.
6. What you can do with the result
The result is rendered as a readable structured report instead of raw model JSON.
Current actions include:
- share the text report
- generate and share a poster
- save the result as a memo
When saved as a memo, it is created as a PRIVATE memo and pushed into the sync queue.
7. What else appears in the report flow
Evidence references
When enough evidence exists, the report shows a reference section:
- expand references;
- inspect the cited fragments;
- jump back to the original memo.
History
Every completed AI insight is stored in history.
You can reopen past reports without generating them again.
Stale warning
If source notes changed after the report was generated, a historical report may show a "notes updated" warning so you know it may no longer reflect the latest state.
8. Common failures and checks
Analysis cannot start
Check these first:
- Is a working
Chatmodel configured? - Is the current card's prompt template empty?
- If this is the custom card, are title, description, and prompt all filled in?
"No content to send"
This usually means filtering left no eligible memos, not that the AI service is down:
- expand the time range;
- include the visibilities you actually need;
- sync first so the local DB has current data;
- confirm memo bodies are not empty.
The run works, but quality feels unstable
Common reasons:
- no embedding model is configured;
- the embedding model is failing repeatedly;
- the chosen insight card is not a good fit for your goal;
- the prompt template is too vague.
Timeout or network error
Check:
- whether
Base URLis reachable; - whether
API Keyis valid; - whether model key is correct;
- whether your gateway requires extra headers.