Journal (freeform and guided reflection)
A personal-reflection space inside Omniscio that opens on Today: one question written for your life today and a box you can type into at once, with freeform writing and guided topics one click away. Replies come from the AI service you pick — your Claude subscription by default, or your own AI service. Entries are stored on your machine.
On this page
What it is
Journal is a virtual project in Omniscio for personal reflection. It opens on Today (below), and supports two modes:
- Freeform — a simple text editor with title, body, auto-save (1.5 s debounce), and tags. A live word count and estimated reading time are shown in the toolbar. Write whatever comes to mind.
- Guided — a Socratic chat with an AI mentor that follows a structured template. The templates live in AI Coaching's shared prompt library: Gratitude, Problem Solving and Emotional Processing are offered under Choose a topic, and the Coach and Daily Reflection templates are folded into Today. Each template shapes the guidance that steers the mentor's questioning style. Replies come from the AI service you pick: your Claude subscription by default (best results, no per-message cost) or your own AI service (the AI keys in Settings → My API Keys, which may cost per message). Journal never switches between the two on its own, and the chat always says which one is answering. You also pick how replies sound with Reply style (Settings → AI Coach): Calm by default (warm and unhurried), Upbeat (encouraging, notices your wins) or Minimal (brief and plain). The style changes only the voice, never the layout or what is sent.
Entries are stored locally in the SQLite database. Your message is saved the moment you send it, so a reply that fails or never arrives never costs you your words. Just Write (freeform) entries can be marked as favorites; a guided conversation has no favorite button in the app (an approved agent can still set the flag through the CLI). Entries can be filtered and sorted in the sidebar, favorites included.
Where to find it
Where it lives
Journal is opt-in and off by default. Turn it on in Settings → Lab → Built-in apps → Journal (setting journalEnabled); it then appears as a virtual project in the left sidebar (amber NotebookPen icon). Gate enforcement exists at two layers — IPC handlers (assertJournalEnabled() wrapping every handler) and CLI routes (the journalReady() preamble returning 403) — and both refuse until the toggle is on, because the unreleased-feature registry holds Journal at status: 'in-development' deliberately.
How it behaves
How to use it
- Open it. Click Journal in the left projects sidebar. It opens on Today (see Today below), with your recent entries under it. The layout splits into a sub-sidebar (entry list) and a content pane, following the same pattern as AI Coaching. If the journal cannot be loaded, the sub-sidebar and the Recent entries section both say so and offer a Retry button that loads it again, instead of showing an empty list; the Today card above it stays, so you can still write today. The landing page's Recent entries always lists your five newest entries — whatever sort or filter the sidebar is using — and it refreshes after every change, including an Undo from the delete toast. The list shows 50 entries at a time; Load more adds the next 50, up to 500, and past that a note points you to search and filters. An entry opened from search, or one outside the current filters, still opens and saves. Entry not found. appears only when the entry has really been deleted, and if it cannot be read right now the panel offers Retry. While the journal is loading, the list and the dashboard show placeholder rows, never "No journal entries yet".
- Create an entry. Today is ready to type into, and the + button in the sub-sidebar header brings you back to it. Under the Today card, Just Write opens a blank page (it saves nothing yet: the entry is created the moment you first write something — a title or body — and leaving a blank page writes nothing), and Choose a topic lists the guided topics (Gratitude, Problem Solving, Emotional Processing) with their emoji icon, name and description.
- Freeform mode ("Just Write"). Type a title (optional) and body. Changes auto-save after 1.5 seconds of inactivity; pending saves flush on unmount so nothing is lost when switching entries. A live word count and estimated reading time appear in the toolbar. Clearing the title saves an untitled entry, and the toolbar shows Saved only when the stored entry matches what you typed. The title holds up to 500 characters and the body up to 100,000, the most a save accepts; typing or pasting past either stops at the limit, so autosave keeps working on a very long entry. Saving… and Saved belong to the entry being saved: open another entry while a save is still going and that entry shows neither. A save made from an out-of-date copy of the entry (another window or an agent changed it first) is refused rather than overwriting the newer text; the sidebar's refresh picks up the current copy and the next edit saves on top of it. While a Just Write page is still blank, the favorite, tag and delete buttons are hidden; they appear once the entry exists.
- Guided mode (pick a template). Choosing a template opens its chat as a draft: nothing is saved yet, so walking away leaves no empty entry behind. The entry is created the moment you send your first message, already holding it and, before it, the opening it answers as the guide's first turn — whatever opening is on screen at that moment. The opening prompt is the template's own. Type your response; Enter starts a new line, and Ctrl+Enter (Cmd+Enter on a Mac) or the Send button sends. The guide answers with Socratic follow-ups; while it thinks that entry shows Reflecting..., which becomes Taking longer than expected... about 5 seconds after you sent it, even if you opened something else and came back. Only the entry whose reply is pending shows it and blocks Send; you can open and write in other entries meanwhile. A reply that could not be fetched leaves its notice, and the button that fixes it, on that entry alone: a reply arriving in another entry never clears it. The full conversation is persisted as
guidedMessageson the entry.- Your words are kept. Your message is saved before the AI is asked. If no reply comes back — the service failed, timed out, or the app closed mid-reply — the message stays in place with the reason in one plain sentence (or, when the reason is no longer known, "I couldn't reply just now. Your words are saved. Try again when you're ready.") and a Try again button that asks again without retyping. The reason shows the same way when Today asked for the first reply, and again when the entry is reopened. If a send itself could not be saved, a message says so and the words stay in the box. Where a sign-in or an AI key is the fix, an Open Settings button sits beside it: it opens Settings → Accounts for a rejected Claude sign-in and Settings → My API Keys for a key that did not work.
- Which service is answering. A line above the input says Replies from your Claude subscription or Replies from your own AI service · For best results, use your Claude subscription, with a Change button that opens the setting.
- A half-typed reply is kept. What you type in the reply box is kept on this computer for that conversation only, filed under your signed-in account: open another entry and its box is empty, come back and your words are still there, even after closing the app. A template you haven't started yet keeps its own. If a send fails, the words go back into the conversation that sent them, never into one you moved to while it was sending.
- Not signed in to Claude. When the Claude subscription is picked and nobody is signed in, a calm notice at the top says so, with a Sign in to Claude button; you can keep writing and everything you send is saved. While the chat stays open it notices when you sign in — from Settings, or from a terminal once you come back to the window — and a reply that failed for this reason then shows Try again, which asks for the reply to the saved message.
- Limits. The replies left for the entry show above the input. Up to 20 replies an hour per entry and 100 a day, counted only when a reply actually arrives; at the limit the input stays usable: what you send is saved to the entry with no reply, and one line says your words are saved and when replies come back.
- Which key sends. In the Journal Enter always starts a new line, whatever the app-wide send key is set to; Ctrl+Enter (Cmd+Enter on a Mac) or the Send button sends. While a Japanese, Chinese or Korean keyboard is still choosing a word, Enter only confirms that word, so it never sends half a reply; adding a tag works the same way.
- Screen readers. Each message is read with who wrote it first ("You:" or "Journal:"), and the guide's emoji is skipped.
- Favorites. In a Just Write entry, click the heart icon in the editor toolbar to mark it as a favorite. A guided conversation has no heart, so it cannot be favorited on screen (an approved agent can still set the flag with
POST /journal/entries/:entryId/toggle-favorite). Favorited entries show a heart indicator in the sidebar entry list. Click the heart icon in the sidebar header to filter to favorites only. - Sort & filter. Click the sort icon (↕) in the sidebar header to choose between Newest first, Oldest first, or Recently updated. Each row's date is the day the entry was written, except under Recently updated, where it is the day it was last edited, so the dates read in the list's own order. Click the calendar icon to reveal date-range inputs that filter entries by creation date. Both ends are Journal days, read in the Journal's time zone like Today, and both are included, so a one-day range shows that day's entries. A date applies only once it is complete: while you type a year the list does not change, and leaving the box half-typed puts back the date that is applied.
- Tags. Click the tag icon in the header to add tags (Enter to confirm). Tags are normalized to lowercase and deduplicated. Click a tag in the sub-sidebar to filter the list to it; the filter covers the whole journal (not just the entries already loaded) and ignores letter case. Tags are listed in alphabetical order for your language, so accented and non-Latin tags sit beside their base letters. When there are more than eight tags, a +N button shows the rest and Show fewer hides them again.
- Search. Click the search icon in the sub-sidebar header. Searches title and content with a 250 ms debounce. Minimum query length is 2 characters, or 1 when it is a Chinese, Japanese, Korean or Thai character, since one such character (猫, "cat") is often a whole word. Every word must appear. Symbols and emoji (
:),--) and words inside Chinese, Japanese, Thai or Korean text are found too. Results load 50 at a time with Load more; while more exist, the count reads "More than N entries found". The results come back in the order the list is sorted (Newest, Oldest or Recently updated), honour the favourites and date-range filters that are switched on, and update when you change the sort or a filter. An open search also refreshes when an entry is added, edited or deleted. Closing the search bar (or clearing it), or leaving the Journal, ends the search and brings the whole list back; a keystroke still waiting out the debounce is dropped, so a cleared search never re-runs. The search box takes at most 500 characters, the longest search the app accepts. If a search fails, the list says "Search failed — please try again." instead of "No matching entries", so a failure never looks like an empty result. - Writing stats. Click the bar-chart icon in the sidebar header to view a stats dashboard. It leads with the daily habit: your current streak and how many of the last 14 days you checked in, where a day counts once you answer its Today question (free writing alone does not, and a missed day quietly ends the streak). Then come total entries, total words, your longest streak, and average words per entry. Words count your own writing — the entry text plus your answers in a guided conversation, never the coach's replies — and the average is taken over the entries that hold any words. Below the cards, Journal health shows the last 14 days: how many days had an entry, how many Learned notes you removed (Undo takes one back off), and the median time per check-in, from opening Today to the first reply. A check-in where you answered the plain "What's on your mind today?" line has no time, because it is only created when you answer. Replies rated "knows me" or "generic" is a line you keep yourself during the two-week trial; the app does not measure it. When the numbers cross one of the trial's limits, a short amber note appears under them: more than 1 in 5 of the notes the Journal learned were removed (it may be saving too much), or a typical check-in takes more than 3 minutes (it may be asking too much). The notes wait for enough to go on: at least 5 notes learned, or at least 3 timed check-ins.
- Export. The export menu in the sidebar header downloads JSON or Markdown of exactly the entries the list is showing, in the order it is sorted (newest first, oldest first or recently updated) — the tag, favourites and date-range filters all apply, and while a search is open (two or more characters, or one Chinese, Japanese, Korean or Thai character) the file holds only the entries that match it, from every tag, just as the search list shows them (the picked tag sits out of a search). The file name carries today's local date, and an entry with no title is headed with the translated word for "Untitled". The file holds every entry the list selects, however many there are: past 500 the app reads them 500 at a time and still saves one file. The JSON file is one object —
truncated(false),totalCount(how many entries the file holds),nextOffset(null) andentries. An agent'sGET /journal/exportstill answers one page of at most 500 entries: a cut-off page says so (the Markdown opens with a notice; the JSON carriestruncated: trueand thenextOffsetwhere the rest start), so a page never passes for the whole Journal. - Delete. Click the trash icon, then confirm. Entries are soft-deleted (
is_deleted = 1). An undo toast appears after deletion — clicking Undo restores the entry immediately, tags included (a deleted entry's tags are hidden, not removed, until the trash purge deletes the entry). If the restore itself fails, a toast says so and the entry stays deleted. Deleting a freeform or guided entry also removes the unsent text this computer kept for it (its local copy and any half-typed guided reply), so Undo brings the entry back without them. Deleting an entry also moves the facts only that entry taught to Recently deleted for 30 days (a fact another entry or a Coaching chat also taught stays), and Undo brings the entry and those facts back together. The confirm says you can restore it from Recently deleted for 30 days (on a paired phone, "on your computer", because the list is on the computer only and the phone does not show it). The deleted entry waits in Recently deleted at the foot of the Journal sidebar on the computer, with AI Coaching on or off: click it to open the list of deleted entries and removed Learned notes, each with the day it is erased for good, and Restore brings an entry back with those facts (or a removed fact back on its own). The same entry also shows in AI Coaching's Recently deleted list as a Journal entry, and when its time is up the entry and its facts are erased for good. Entries deleted before the change keep their 90 days; deleted after, 30. - Learned. When you leave an entry, or leave it untouched for 10 minutes, the Journal asks AI Coaching's one fact-learning step what is worth learning from your own words (never the coach's replies or the opening question, at most 5 facts an entry). The facts show in a Learned row at the foot of the entry — a bookmark, the word Learned, and one chip per fact on a faint accent tint — every time the entry is opened. Hover a chip to read the whole fact; a fact that replaced an older one says which ("replaces: …"). Click a chip's × to remove that fact: a toast offers Undo, and keyboard focus moves to the next chip (or, after the last one, to the reply box or the entry text). Editing an entry re-checks its facts with no AI call: a fact whose sentence you deleted goes, and putting the sentence back does not bring it back. A fact another entry or a coaching chat taught first stays when you edit a later entry that said it again. A fact you removed is never learned again. Facts live in AI Coaching's shared memory, tagged as from the Journal; they work with AI Coaching switched off. If learning fails twice in a row (the AI is down, out of budget or errors), one inbox card says Journal couldn't learn from your entries; your entries stay saved, recent ones are tried again the next time Omniscio starts, and the card clears itself after the next good learn.
Today
Today is the Journal's landing screen and the first thing you see each time you open it.
- Today's question. One question written for your life today, from what the Journal already knows about you: about two weeks of your entries and, while AI Coaching is on, your profile and key facts. Questions rotate across five kinds — gratitude, a person, a goal, a worry and a win — and avoid what was asked in the last two weeks. It is written once a day, about half an hour before your nudge, so it is already waiting when you open Today; opening Today again that day shows the same question without asking the AI again. Left open past midnight, the card moves to the new day by itself. Nothing is written ahead of time until you have opened Today once (or written something), so nothing goes to an AI before your first visit. After three nudges in a row with nothing written, nothing is written ahead of time until you next open Today or write something, and Journal never writes ahead while guided replies are off. After a gap of seven days or more since you last wrote, the question becomes a gentle catch-up about what has changed; it never says how long it has been, and if it cannot be written, a short welcome-back line shows instead.
- You can type at once. "What's on your mind today?" shows straight away. If today's question is ready within 5 seconds and you have not started typing, it takes that line's place. Once you type, the question never changes under you, and words kept in the box from your last visit count as typed. With guided replies off, no sign-in, or nothing known about you yet, the plain line simply stays, with no error.
- What is sent. While guided replies are on, one plain line above the answer box (and above the reply box in a guided reflection) says what your AI sees to write a reply: the conversation, about two weeks of your recent entries and, with AI Coaching on, your profile and key facts, and nothing else. It is hidden while guided replies are off, because nothing is sent then.
- The first visit. A brand-new journal starts with three quick questions to get to know you, one at a time in the Today box: what's going on in your life right now, who matters most to you, and one thing you're working toward. Each has a Skip link, and a small line says which of the three you are on. Each answer is saved as soon as you send it. After the third, the Journal replies to what you shared, and from then on the questions can be about your life. Skip all three and you get the plain opener instead. Once you finish or skip it, it does not come back on that computer or phone.
- Different question. Swaps the question for a new one, up to three times a day; after that the button is disabled and says "Three is the limit for today". If a new question cannot be written, it brings back one you were shown earlier that day. Swapping never uses up your reply budget, and it is not offered once you have answered, or while "Send guided writing to an AI for replies" is off (only an AI can write a question).
- This hit / Missed. Under a personal question, two small buttons let you say whether it landed. Click the other one to change your mind; you can rate before or after answering. Later questions skip a kind you marked Missed and lean toward the ones that hit, and the Journal health section in Stats shows how many rated questions hit over the last two weeks. Rating sends nothing to an AI, and the plain "What's on your mind today?" line has no buttons.
- Answering. Type your answer (Enter makes a new line; Ctrl+Enter, Cmd+Enter on a Mac, or Send sends). Your words are saved into today's entry under the question you actually saw, before any AI is asked; then the entry opens and the reply arrives as in any guided conversation. Once today's question is answered, the card offers Continue today's entry instead of a second box — unless words are already waiting in the box (say you answered in another window first): then the box stays, and Send adds them to today's entry.
- An unanswered question is not an entry. Until you answer, today's question does not appear in the entry list, search, export or stats.
- Just Write and Choose a topic sit under the card for a blank page or a guided topic instead. Stats live in their own view, not on this screen.
- The daily nudge. Once a day, at your nudge time (8:00 PM unless you change it), one inbox card, Time for your Journal, opens Today. If Omniscio was closed or the computer asleep at that time, the card shows once when Omniscio next opens that same day. A card you leave unanswered is put away when the next day starts, and while the Journal is on it is the only daily card: the AI Coaching daily check-in card steps aside.
Go deeper
When one entry needs real thinking, Go deeper continues it in AI Coaching.
- Where it is. A secondary button under the latest reply of a guided entry, and below the text of a written entry. It is not on the phone.
- One line first. It asks "Continue in AI Coaching? The coach will read this entry." Confirm, and the Coaching session opens. If AI Coaching is off, the same dialog says so and offers Turn on; your entry is untouched either way.
- Nothing to repeat. The session starts with your entry as your own first message, so the coach answers what you wrote. Nothing from the entry is written into the coach's working file, and it works with the "coach can see my entries" switch off, because you are handing over this one entry yourself.
- Linked both ways. The entry then reads Continued in AI Coaching and reopens that same session. If that session is gone, or AI Coaching is off, it asks first with the same one-line dialog instead. The session is the Go Deeper interview, also a card under Journaling in AI Coaching, whose write-up, Journal Deep Dive, keeps every earlier deep dive as a short dated entry.
- When it can't. If Go deeper is refused after you confirm (the AI switch was turned off, too many starts in a minute, or the coach could not take what you added), the dialog closes and a short message says why, for example "Couldn't send what you added to the coach just now. Your entry is saved. Try again." Your entry is untouched.
- Erased together. When the entry is erased for good (30 days after you delete it), the copy of it in that session goes too: your first message there now reads "This Journal entry was erased.", it is gone from search and from the session's saved summaries, and the session's own transcript is rewritten the same way. So does each message that Continued in AI Coaching sent later with words you added or changed ("I added this to my journal entry:" …). The coach's replies and anything you typed in the session yourself stay. While the coach is in the middle of a reply, the erase waits for a later day.
Where guided replies come from
Open Settings → Plugins → Journal and find the replies-come-from picker (it shows while guided replies are on). It is the same setting as Your AI service in Settings → AI Coach, which shows while AI Coaching is on: change it on either page and both follow. Journal replies and today's question use it, and AI Coaching moves onto it next. Pick:
- Your Claude subscription — best results (recommended) — the default. Replies use the Claude login you already have, with no per-message cost. If you are not signed in, the chat says so and offers Sign in to Claude.
- Your own AI service — may cost per message — replies use the AI keys in Settings → My API Keys (for example OpenRouter or DeepSeek). If no key works, the chat says so and offers Open Settings, which opens My API Keys at the key order list.
Today's question and the facts the Journal learns come from the same service. Journal never switches on its own: if the picked service cannot answer, the chat tells you why and keeps your message.
Three switches
Every Journal setting lives on one page, Settings → Plugins → Journal: the two switches below, Reply style, the nudge time and the replies-come-from picker. AI Coach's settings carry none of them.
- Journal itself (Settings → Optional Features). Turning it on keeps your writing on this computer.
- Send guided writing to an AI for replies (Settings → Plugins → Journal, on by default). Off means nothing you write leaves this computer: no facts are learned, a guided message still saves but gets no reply, the chat shows Guided replies are off with a Turn on in Settings button, and Today shows its plain opener instead of a written question. The replies-come-from picker is hidden while it is off.
- Let the AI Coaching coach see my recent entries (Settings → Plugins → Journal, on by default). Off means an AI Coaching session is refused at the Journal, so the coach never sees an entry and no approval card is raised for it, and the Journal learns no new facts from your entries: nothing is sent to learn from and nothing is saved, and an entry you finish while it is off is never learned later, even after you turn it back on (an edit you make once it is on is). Facts learned earlier stay on their entries. The switch says so under its name. Other agents keep the ordinary approval.
An install that never touched the two Journal switches behaves exactly as before they existed. Turning either off deletes nothing. The two switches and the replies-come-from picker change on the computer only: on a paired phone they are greyed out with "You can change this on your computer."
Settings search finds the two switches only while Journal is on, and finds the "replies come from" picker only while guided replies are on — the same rows the AI Coach section draws.
The daily nudge on your phone. The Journal's daily card can also ring a paired phone. It is off until you choose it: Settings → Notifications → Alert types, find Journal daily nudge, and turn on Phone. The nudge arrives from your first day with Omniscio.
Cost and privacy
- Freeform mode makes no AI calls of its own, apart from learning facts.
- Learning facts sends your own words from a finished entry to the same service once (one extra call per finished entry, only while Send guided writing to an AI for replies and Let the AI Coaching coach see my recent entries are both on). An entry you open again without changing your words is never sent twice.
- Guided mode sends the template's guidance and the most recent 20 messages of the conversation to the one service you picked, on each reply, together with a short background about you so the reply can be personal: about two weeks of your journal entries (up to five, each shortened) and, while AI Coaching is on, your coaching profile and key facts. It never sends your whole history. On your Claude subscription it runs as a single tools-off Claude request on your own login: no per-message charge, though replies count toward your plan's usage limit. On your own AI service it is a normal paid call, recorded in the spend log like any other. Replies are capped at 20 an hour per entry and 100 a day.
- The conversation is kept only in the Journal entry — never copied into a session, the inbox or search. On your Claude subscription, guided replies and today's question ask Claude to save no log of the run on your computer, and they run without your own Claude Code add-ons (hooks, skills, plugins, MCP servers and CLAUDE.md files), so none of them sees what you wrote. A log Claude Code saved before this change stays where it is until you remove it.
- All journal data is stored in the local SQLite database, and Journal itself syncs nothing. Your entries still leave the computer in three places: guided replies and today's question are written by your AI from that same short background (only while Send guided writing to an AI for replies is on); an AI Coaching session reads your recent entries live from the Journal, only while Let the AI Coaching coach see my recent entries is on and only after you approve that session with one click (the coach's own working file holds no entries); and a backup includes the journal tables, so your journal goes wherever your backups go.
- AI agents on this computer can read, change or delete your entries, through the Journal routes of Omniscio's local control server. Each agent needs your one-time approval first, using its own session key (see Agent access below); the computer's shared full-access token is refused.
Related
- AI Coaching — structured interview sessions with artifact capture
- Daily Journal Check-In — AI Coaching check-in nudges (separate feature)
- Journal — part 2 — the database, list and search, and the design reasons behind them
- Journal — part 3 — the file inventory, IPC channels, guided mode internals and control-server routes
Last verified 2026-11-11