Learn · AI organization
AI proposes the plan; you approve every item
Echo Notes never lets AI rewrite your notes silently. When you classify a single note or a batch of up to 100, the assistant first presents a plan showing current and proposed folders, progress, and any failures. You inspect every item and accept only what you approve; approved changes apply atomically to local content. The latest supported classification can be undone when no content conflict exists.
01 · Definition
Reviewable AI, not automatic AI
Reviewable AI means every AI-generated note or automation draft is presented with its title, body, destination, and impact before anything is created. For classification, the plan lists each note’s current folder next to the proposed folder so you can spot mistakes before approving. Nothing is applied until you say so, and partial acceptance lets you approve the good items while skipping the rest.
The same approval rule covers the conversational assistant described on the Features page: drafts first, creation only after your decision.
02 · Steps
Classify a batch without losing control
Use this routine whenever unfiled notes start piling up.
- Your problem
- Dozens of captured notes need structure, but a broad, irreversible AI rewrite would be too risky.
- Product behavior
- Echo Notes shows the full classification plan with current and proposed folders, progress, and failures, with partial acceptance and failed-batch retry.
- Your result
- The scope stays visible, and only approved items are applied atomically to local content.
Select the scope
Choose a single note or a batch of up to 100 notes to classify.
Inspect the plan
Review each proposed move, keep the correct ones, and skip anything doubtful.
Apply and verify
Apply the approved items, check the result, and undo the latest supported action if it was wrong.
03 · Criteria
A checklist for trusting an AI plan
| Check | What good looks like | If it fails |
|---|---|---|
| Scope | Only the notes you selected are listed. | Narrow the batch and regenerate. |
| Destination | Each proposed folder matches the note’s topic. | Skip that item; approve the rest. |
| Failures | Failed items are labeled, not silently dropped. | Retry the failed batch. |
| Reversibility | You know undo covers the latest supported action. | Approve a smaller batch first. |
04 · Limitations
What approval does not promise
Approval before writing does not remove the need to process the request online: AI features require a network round trip to the configured model provider, available account usage, and gateway capacity. Undo applies to the latest supported note creation or classification action when no content conflict exists; it does not mean every AI or automation operation is reversible. Your notes and conversations are not used to train foundation models, and only the content needed for a request is sent to the provider. Details are in the Privacy Policy.
05 · Echo Notes practice
Organize this week’s captures
Select up to 20 unfiled notes and request classification. Approve five items, skip the rest, and confirm the applied moves in their folders. Once the structure feels right, explore cloud automation basics for recurring checks, or compare approaches in Echo Notes vs Notion. Plan differences that affect AI usage are listed on Pricing.
FAQ
AI organization questions
Does AI change my notes automatically?
No. AI-generated drafts and classification plans require your approval before anything is created or applied.
Can I approve only part of a plan?
Yes. Partial acceptance is supported, so you can approve correct items and skip doubtful ones, then retry failed batches.
How many notes can I classify at once?
You can classify a single note or select a batch of up to 100 notes per classification run.
Can I undo an AI classification?
The latest supported classification can be undone when no content conflict exists. Not every AI or automation operation is reversible.
Is my content used to train AI models?
No. Notes and assistant conversations are not used to train foundation models; providers process only what each request needs.