187 lines
6.3 KiB
Markdown
Executable File
187 lines
6.3 KiB
Markdown
Executable File
# How Memories Work
|
|
|
|
AdventureOS memories are durable facts the mentor can use as context in future replies. They are separate from raw chat logs, reflections, and adventure history. A saved memory has a category, title, content, source, and enabled/disabled state.
|
|
|
|
The fastest way to make the mentor register a memory request is to use direct save language at the start of your message.
|
|
|
|
## Best Phrases To Use
|
|
|
|
Use one of these phrases, then state the fact clearly:
|
|
|
|
- `Remember that ...`
|
|
- `Remember this as ...`
|
|
- `Save this ...`
|
|
- `Save this as ...`
|
|
- `Add this to memory ...`
|
|
- `Add this as ...`
|
|
- `Add a memory ...`
|
|
- `I want you to add a memory ...`
|
|
- `The memory will be ...`
|
|
|
|
Good examples:
|
|
|
|
- `Remember that I prefer short explanations.`
|
|
- `Save this: I get discouraged when my task list is too large.`
|
|
- `Add this to memory: I like learning with worked examples.`
|
|
- `I want you to add a memory for me, and the memory will be a goal type: One of my goals for the next few weeks is to wake up at 7am every morning.`
|
|
|
|
## Tone
|
|
|
|
You do not need a special tone. Be explicit, plain, and declarative.
|
|
|
|
Strong memory requests usually have three parts:
|
|
|
|
- A save phrase: `remember`, `save`, or `add this to memory`
|
|
- A complete fact about you
|
|
- Enough detail that it will still make sense later
|
|
|
|
Less reliable:
|
|
|
|
- `That's useful for later.`
|
|
- `You should know this.`
|
|
- `Keep that in mind.`
|
|
- `Maybe remember this somehow.`
|
|
|
|
Those may make sense to a human, but the app is looking for direct memory intent.
|
|
|
|
## What Happens After A Request
|
|
|
|
When you send an explicit memory request in mentor chat, AdventureOS first tries to extract a memory candidate from your message.
|
|
|
|
Explicit save requests do not require memory learning to be enabled. They use the memory action path directly.
|
|
|
|
If approval is not required, the memory is saved immediately and the mentor should say something like:
|
|
|
|
`I've saved that as a goal memory: "Wake Up at 7am Every Morning".`
|
|
|
|
If approval is required, or if the category is sensitive, the app creates a pending suggestion instead. The mentor should say something like:
|
|
|
|
`I've created a memory suggestion for review: "Wake Up at 7am Every Morning".`
|
|
|
|
A suggestion is not confirmed memory yet. You can accept, edit, reject, or ignore suggestions in `Settings -> AI Memory -> Suggestions`.
|
|
|
|
## Approving Or Rejecting In Chat
|
|
|
|
If the mentor has just created a pending memory suggestion in the same chat, you can approve it with:
|
|
|
|
- `approved`
|
|
- `yes approved`
|
|
- `save it`
|
|
- `confirm it`
|
|
- `accept it`
|
|
- `that is correct`
|
|
- `that's correct`
|
|
- `yes that's right`
|
|
- `yes that is right`
|
|
- `yes, as a goal, approved`
|
|
|
|
You can reject it with:
|
|
|
|
- `reject it`
|
|
- `don't save that`
|
|
- `do not save that`
|
|
- `forget that`
|
|
- `ignore that suggestion`
|
|
- `no, don't remember that`
|
|
- `no, do not remember that`
|
|
|
|
Editing a pending memory directly in chat is currently limited. If you say `change it to`, `edit that`, `update that suggestion`, or `make it`, the mentor should tell you to edit the pending suggestion card before approving it.
|
|
|
|
## Passive Memory Suggestions
|
|
|
|
AdventureOS can also suggest memories from ordinary messages when memory learning is enabled. This is different from explicit save requests.
|
|
|
|
Examples that can become suggestions:
|
|
|
|
- `One of my goals for the next few weeks is to wake up at 7am every morning.`
|
|
- `I like learning with examples.`
|
|
- `I dislike too many tasks at once.`
|
|
- `I get worried when I fall behind.`
|
|
- `I prefer concise explanations.`
|
|
- `I want the AI to be direct with me.`
|
|
- `I struggle with planning my week.`
|
|
|
|
Passive learning depends on `Settings -> AI Memory -> Learning`:
|
|
|
|
- `Enable memory learning` must be on.
|
|
- `Suggest after reflections` must be on for reflection-based suggestions.
|
|
- `Allow learning sensitive categories` controls sensitive categories.
|
|
- `Require approval before saving` decides whether suggestions must be reviewed before becoming confirmed memories.
|
|
|
|
The underlying settings also include chat-based suggestions. By default, chat suggestions are enabled when memory learning is enabled, but explicit save requests are still the most reliable route when you definitely want a memory created.
|
|
|
|
## Categories The App Understands
|
|
|
|
Memories are grouped into categories:
|
|
|
|
- Identity & background
|
|
- Long-term goals
|
|
- Current goals
|
|
- Likes
|
|
- Dislikes
|
|
- Motivators
|
|
- Things that discourage
|
|
- Daily routines
|
|
- Weekly routines
|
|
- Spiritual practices
|
|
- Reading preferences
|
|
- Learning interests
|
|
- Exercise preferences
|
|
- Work/study commitments
|
|
- Worries & concerns
|
|
- Preferred AI tone
|
|
- Boundaries & avoid
|
|
- Important personal context
|
|
- Current life season
|
|
- Open questions about user
|
|
|
|
If your wording mentions goals, current goals, trying to do something, or wanting to do something, the app will usually classify it as `Current goals`. If your wording says you prefer a kind of explanation, it will usually classify it as `Preferred AI tone`. If your wording says you like learning a certain way, it may classify it as `Learning interests`.
|
|
|
|
## Manual Control
|
|
|
|
You can bypass chat phrasing entirely in `Settings -> AI Memory -> Memories`.
|
|
|
|
From there you can:
|
|
|
|
- Add a memory manually
|
|
- Search memories
|
|
- Filter by category
|
|
- Disable or re-enable a memory
|
|
- Archive a memory
|
|
- Export memories
|
|
- Reset all memories
|
|
- Edit the profile summary used as compact mentor context
|
|
- Rebuild the profile summary from saved memories
|
|
|
|
## Practical Recipes
|
|
|
|
For a goal:
|
|
|
|
`Remember that one of my current goals is to read for 20 minutes every evening.`
|
|
|
|
For tone:
|
|
|
|
`Remember that I prefer direct explanations with one example.`
|
|
|
|
For a worry:
|
|
|
|
`Save this: I get worried when I fall behind and need help choosing the next small step.`
|
|
|
|
For a boundary:
|
|
|
|
`Add this to memory: I do not want shame-based motivation.`
|
|
|
|
For a learning preference:
|
|
|
|
`Remember that I like learning with examples before theory.`
|
|
|
|
## Troubleshooting
|
|
|
|
If the mentor does not register the memory request, rewrite it with `Remember that ...` or `Save this: ...`.
|
|
|
|
If the mentor says it created a suggestion, go to `Settings -> AI Memory -> Suggestions` and accept it.
|
|
|
|
If the mentor says there is no pending suggestion to approve, the suggestion was not created in the current chat, was already handled, or the approval phrase was too detached from the original request.
|
|
|
|
If a memory already exists, AdventureOS avoids creating a duplicate and should tell you that the memory is already saved or pending review.
|