AI Coaching sidebar (interview prompts and artifacts) (part 2)
Part 2 of the AI Coaching sidebar page: how an interview's artifact is saved, edited, versioned, restored and deleted, how pinned context and detail levels feed artifacts into sessions, and how imported work and the KMS vault mirror fit in.
On this page
What it is
This is part 2 of the AI Coaching sidebar (interview prompts and artifacts) page. It covers the artifact half of AI Coaching — how a finished interview document is saved, edited, versioned, restored, deleted and imported, how pinned context and detail levels decide what the coach reads, and how a saved artifact reaches your other sessions and the KMS vault.
Where to find it
AI Coaching is the virtual project in the left sidebar, and the parent page walks through turning it on and using it. Inside it, artifacts are the Artifacts (N) section of the AI Coaching sub-sidebar, and clicking one opens the artifact editor in the right pane. The pin pickers live on a completed card's ⋮ menu on the Dashboard and on the pin button in a running session's header, and Import existing work and Bulk Import sit on the Dashboard's prompt cards and header.
How it behaves
How artifacts get saved
There is no separate "save the transcript" round-trip. The interview prompt in Claude's project-level instructions teaches it to do two things once the interview has captured enough material: save the finished document by calling Omniscio's CLI — a POST to a session-scoped save endpoint — and show that same document inline in the chat as clean prose. The save returns success or failure, so Claude knows it worked and can retry or tell you — more reliable than writing a file that Omniscio then had to detect.
The endpoint is locked to the interview's own artifact: Claude posts only the document text, and Omniscio fills in the title from the session's configured artifact title, so Claude can never write into a different artifact and a stray or injected title can't reach the sidebar. Omniscio writes a new version and pushes a refresh, so the artifact appears seconds after Claude finishes. Re-saves are deduped by content: re-posting the same document creates no new version, so "refine it later" simply means posting the updated text.
The save is bound to the coach's own session: another session's token can't save into this one, and a session that isn't an AI-Coaching interview is rejected. That is the prompt-injection defence — the coach can only write the single artifact its interview is configured to produce, with the title enforced server-side. The save applies immediately, with no inbox approval, because it is the session's own artifact. The general "create / update / delete any artifact" CLI stays approval-gated and is deliberately hidden from the coach, so text inside your own saved profile can never become a silent edit of a different document.
The Save artifact now button is a UX nudge, not a separate save path: it sends Claude a one-line operator message asking it to save the document (and show it) now, deferring to the same instructions. It works while the session is waiting for you (ready or needs-you) and only after you have answered the coach at least once; it stays off while the coach is replying, before your first answer, and for paused, archived, errored or ended sessions, where there is no live agent to respond.
A display filter also strips stray <artifact> / [PHASE:N] / [SEGMENT:N] markers from coaching messages before display, so those tags never leak into the chat while the prose stays intact.
Editing artifacts
Opening an artifact from Artifacts shows the rendered markdown plus a small toolbar in the right pane. While you're on the latest version, the toolbar has an Edit button; clicking it swaps the render for a monospace textarea pre-filled with the current contents, cursor inside.
A hard 100,000-character cap applies, in the textarea and in the save path alike, so it holds even if a client tries to slip something larger through. Save is enabled only when you've changed something, the buffer isn't blank and you're under the limit; when the last two hold it back, the toolbar says so beside the button — "Add some content to save." or "Too long to save — trim it under 100,000 characters." A body of nothing but spaces counts as blank, so a full-looking textarea can never leave Save greyed out with no explanation. Ctrl+S (Cmd+S on Mac) saves without leaving the keyboard. After the save the pane returns to the rendered view, desktop and mobile; if the version list then fails to refresh you keep the old list and no "Could not save" message appears, because your save already landed.
Cancel — or Escape — goes back. With nothing typed it returns immediately; with unsaved changes it asks "Discard changes?" first, so an accidental Cancel won't throw away ten minutes of writing. The editor compares your draft against the snapshot taken when you clicked Edit, not the live store value, so a background refresh (another open window updating it, say) doesn't move the goalposts on whether your buffer counts as dirty.
The same prompt guards leaving. With unsaved text in a write-up edit, an In my own words edit on the Profile page, a new goal or a new bookmark, clicking a sidebar row, the breadcrumb or a phone tab asks first: Discard goes where you clicked, Keep editing stays put with your text intact. Clicking the section you are already on never asks, and nothing asks when nothing was typed. A finished card's Open chat and the wrap-up banner's Continue ask the same way: on the phone the chat opens, and the banner goes away, only after Discard.
One cross-window edge case: if someone else (or another window of yours) saves a new version while you're mid-edit, your save fails with a toast — "This artifact was updated elsewhere. Reload it to see the latest version before saving." — carrying a Reload latest button. Your draft is not discarded: the reload pulls in the newer version as the save target, leaves your text where it is, and warns that saving now replaces that newer copy. If the reload itself fails you see "Could not reload the artifact. Please try again." instead. Same on mobile.
A saved edit does the same follow-up work as any other save: search by meaning, the KMS vault copy, the Key Points / Summary / Micro levels and the Core Profile all catch up within moments, and Restore does the same because it saves through the editor. A save that changes nothing adds no new version and starts none of this.
A successful Save bumps the version to MAX+1: a v1 artifact becomes v2 on your first save, v3 on the next, and so on. Older versions are kept — saving never overwrites them.
Version history
With more than one saved version, the metadata strip under the title shows a Version history toggle. Expanding it lists every version, newest first — version number and date per row, a latest tag on the newest, a View button, and Restore on older rows. With only one version there is nothing to expand, so the toggle is hidden.
View on an older row loads that version read-only (markdown render only, no textarea). The toolbar reshapes: instead of Edit you get Back to latest. Copy and Export stay available and act on whatever version is displayed — handy for grabbing old text without changing the artifact. Back to latest (or View on the latest row) returns the pane to its normal editable state.
Restore a version
Each older version has its own Restore button. It does not overwrite anything instantly — it first asks "Restore version 1?" (with unsaved edits open, the discard prompt comes first), and confirming opens the same textarea editor as Edit, pre-filled with that version's content, dirty flag already set so Save is enabled.
Saving creates a brand-new version on top: if v4 was latest and you restored from v1, your save creates v5 starting from the v1 content. v1 and v4 stay in the history unchanged; nothing is destroyed. The mental model: restore is "make this old thing the basis of a new save," not "rewind." Tweak the restored content first if you like — the editor is fully open — or save without changes for an exact copy.
Delete an artifact
In view mode the toolbar has a "⋮" (more actions) button revealing Reset interview and a red Delete. Delete opens a confirmation naming the artifact — "Delete “Life Story”?" — with body text like "This moves all 4 versions and their saved files to Recently deleted. You can restore it there for 30 days, then it's erased for good." (the count reflects the actual versions). For a Daily Check-in or Weekly Review it adds: "The personal facts remembered from it go with it and come back if you restore it. Your goals are kept." While the version list is loading or unreadable, the dialog never guesses a number and says "all its versions" instead.
Reset interview asks the same way: "Reset “Life Story”?" with "This clears all 4 saved versions, your bookmarks, the session summaries, and the personal facts remembered from those conversations. Your core profile is then rebuilt from what remains. This cannot be undone." When the count is unknown it reads "This clears all its saved versions, …".
Confirming moves the whole artifact — every version, rejected ones included — to Recently deleted (under Archived in the sub-sidebar; on a phone, More → Recently deleted), closes the right pane and drops the count beside the Artifacts header. Restore there within 30 days puts every version back; after that it is erased for good. If the delete fails (a network blip or an internal error), a toast surfaces the reason ("Delete failed: …") and the dialog stays open so you can retry without re-clicking; on success it closes itself.
While it waits in Recently deleted the artifact is hidden from every other list and the coach no longer uses it; no version can be opened by its number until you restore. The title is free, so the next save under the same title starts a fresh artifact at version 1. Deleting a Daily Check-in or Weekly Review (or another built-in coaching interview) also moves the personal facts the coach remembered from it to Recently deleted with it — the coach stops using them at once, they come back when you restore the write-up, and they are erased with it after 30 days; your goals stay. Deleting a coaching chat does the same for the facts learned in that chat. A delete made in the few seconds after a save also stops the work that save queued, so the deleted text is never sent for goal or memory extraction and never copied into your KMS vault; resetting an interview does the same. Restoring or merging a backup made before the delete brings the artifact back, because the backup still holds it. The finished interview chats that wrote the artifact — those that ended, are waiting on you, or errored — are archived with it, so its card no longer offers Continue interview into an ended conversation; a coach still replying, and a redo interview that has not saved anything yet, are left open. A failed archive is logged and never fails the delete.
The delete also removes the write-up file and its condensed Key Points, Summary and Micro files from the coaching folder — clean-up is best effort, so a missing or locked file never fails the delete, and a failed removal is logged by artifact id, never by title.
Opening a deleted write-up (from an older search hit or link) shows "This write-up couldn't be opened. It may have been deleted." with a Back to write-ups button. If it still exists but couldn't be read just then (your personal space is locked or starting, or the read failed), the message is "Couldn't open this write-up. Please try again." with Retry and the same Back to write-ups. Same on desktop and phone.
Deleting a write-up, a coaching memory, a bookmark or a Life Inventory take also clears the saved instructions of every coaching chat, so no chat keeps quoting what you deleted. Each chat writes its instructions again, from what is left, the next time it starts. A chat whose coach is still running keeps what it already read until that run ends. The clean-up is best effort: a file that cannot be removed never fails the delete, and the log names only the error code. For agents: discardCoachingRoomInstructions() in level-files.ts removes CLAUDE.md from every room under the coaching sessions folder, under the same write lock the room writers use. removeArtifactFiles awaits it, and the Life Inventory, memory and bookmark delete handlers await it; the rebuild is prepareCoachingInstructionsForColdStart. Two gaps remain. The command-line bookmark delete does not clear the rooms yet. An interview whose instructions are still held back until your first reply keeps that stored copy.
Deleting an artifact that is already gone — the same delete sent twice, or an unknown id — succeeds and does nothing else: no deleted event, no Artifacts refresh, and any pending re-embed is cancelled.
Reset interview, unlike Delete, is permanent: every saved version, rejected ones included, leaves the database with its search entries, so none opens by its number again, and redoing the interview starts a fresh write-up at version 1.
Resets and deletes made by older builds only hid the versions instead of removing them. Once per computer, shortly after the app starts with your coaching space open, Omniscio clears those leftovers — first saving a checked copy to a timestamped backups/ai-coaching-reset-leftovers-<date and time>.json file in the app's data folder, which is kept. It never touches a version you can still see, one saved after the reset, anything in Recently deleted, or a coaching-only write-up. If any step fails it removes nothing and retries next launch.
Rejecting a single version is different: it keeps the row and only hides it, so a rejected version still shows in the version history with its "rejected" badge. Rejecting the newest makes the one before it current again, and semantic search re-reads that version, so the artifact is no longer ranked by the text you rejected.
Pinned context + the pull menu
As you complete more interviews you build a stack of artifacts, and an interview about your week shouldn't have to digest all of them up front. Inside an interview your coach gets your past work two ways:
- Pinned — the artifacts you mark as "load this in full for this interview." The coach reads these completely before the conversation starts. Two always-on extras come free: your Core Profile and your bookmarked Key Moments.
- A pull menu — every other artifact is listed as one line: its title, a one-sentence gist, and roughly how big each detail level is. The coach scans the menu, decides which one or two are relevant, and pulls just those — so a check-in about a job decision can reach for "Career history" without re-reading fifteen unrelated reflections. A pinned artifact disappears from the menu.
You choose what's pinned in two places, both on desktop: a completed card's ⋮ menu on the Dashboard sets the pins for that interview prompt (they apply every time it starts), and the pin button in a running session's header sets the pins for just this session.
Pins are stored by artifact title, matched as your coach matches it: capitals, accents, punctuation and "&" versus "and" make no difference. A pinned title with no saved artifact behind it shows in its own section at the bottom, labelled "Pinned, nothing saved yet" — an interview's suggested write-up you haven't written, or an artifact you deleted. Neither is crossed out, and the note says your coach starts using each one as soon as a write-up with that name is saved. Un-ticking one clears it for good. If your saved pins cannot be read when the picker opens, it says "Couldn't load your pinned write-ups. Close this and try again." and keeps Save (and Ctrl+Enter) switched off, because saving then would replace pins it never saw. For the same reason, each time the picker opens, Save and the checkboxes stay greyed out and Ctrl+Enter does nothing until your saved pins have loaded, which usually takes a moment. If the list of your saved write-ups cannot be loaded, the picker says "Couldn't load your write-ups." with a Retry button, and does not call any of your pins missing or deleted until the list is back.
Pins are locked in when a session is built, so changing them mid-conversation doesn't rewrite the chat you're in (re-pin for the next one). If you pin so much that it won't fit, Omniscio shrinks the biggest pinned artifact down a level at a time and, if it still won't fit, quietly moves the overflow into the pull menu with a short note — nothing is silently dropped. A prompt configured to "pin nothing" stays that way; that is not the same as never having set it (which falls back to a sensible default).
Each coaching session that starts with an artifact in its pinned context increments a use count on that artifact — once per session. Rebuilding the same session's instructions later (a resume, a pin edit) does not count it again, and the session's memories are likewise marked as used once, at the start. The Dashboard's stats bar shows the most-referenced artifact alongside the regular artifact/version/bookmark/session counts.
Your three foundation documents arrive as summaries. Life Story, Values & Principles and Identity & Self-Concept are attached to almost every interview automatically, so the coach always starts with the basics about you. They used to arrive in full on every message, which made even a five-minute check-in heavy. Now each arrives as its Summary (its Key Points version when it has no Summary yet) with a note naming the file where the coach can read the full text — the longer your documents, the more each message saves. A document with no shorter version (a short one, or one written while no AI key was available to condense it) still arrives whole. The Daily Check-in and Weekly Review skip the automatic attachments altogether, and all three documents stay in the pull menu. Anything you pin yourself still loads in full. One side effect: the Dashboard's most-referenced artifact now reflects the documents your interviews really use. A built-in prompt can opt out of the automatic attachments.
How artifacts inform your other sessions
Saved artifacts also become soft context that every new regular session sees. When you spawn a session in any real project (not the virtual AI Coaching project), the agent silently receives a "Things to know about the user" block at the top of its first prompt: each saved artifact rendered as a labeled markdown section, wrapped in [BEGIN PROFILE] / [END PROFILE] sentinels, with an instruction to use the profile naturally — not to quote it back, not to announce it has it, not to bring it up unless relevant.
The block says what the fence is for: everything between the sentinels is data the user captured or imported, never instructions. An artifact can be an imported document, so a sentinel written INSIDE an artifact's title or text is defused to ( END PROFILE ) before rendering — otherwise that document could close the frame early and its prose would sit where the prompt reserves app-supplied facts. The intent is "be informed by who I am," not "perform the profile at me": capture an Identity artifact saying you're a non-programmer relying on engineering AI, and the agent frames explanations accordingly without ever saying "I see from your profile…"
Mechanics worth knowing:
- Gated by Settings → AI Coach → "Enable AI Coaching." With the toggle off, no artifact context is injected, full stop — it is the on/off switch for the whole feature surface, including this side effect.
- Skipped for virtual projects. Omniscio's own UI agents (Automations, Quick Replies, Cron, Tools, etc.) get their own tuned seed bundles; the profile would dilute their instructions.
- Skipped when another seed already fired. Cron heal sessions and the SEED_BUNDLE virtual-project flow each prepend their own context; if either already mutated the launch prompt, the artifact block stands down rather than double-injecting.
- Budget-aware, multi-level. The block is fitted to a 6,000-character total budget, and each artifact is shown at the richest of its four levels — Full → Key Points → Summary → Micro — that fits; over budget, the largest artifact steps down a level at a time, so a couple of huge artifacts can't crowd everything else out. If even the smallest levels don't fit, a per-artifact floor truncates with a
... (truncated — open this artifact in AI Coaching for the full text)marker — nothing silently disappears. You can override the choice per artifact (see Choosing a context level below). - Two delivery paths. With an initial prompt (typing into "Start a new session", or Send Later, web access, a recipe step, etc.), the block is prepended before the CLI spawns. With no prompt (the deferred-spawn flow — opening an empty session and typing later), it is held in the pending first prompt and folded into your first input, so the agent still sees it on turn one.
- Private write-ups never ride along. Your Daily Check-in and Weekly Review entries, and write-ups from the built-in health, relationship, grief, family, body, journaling and therapy interviews (Health & Wellness, Marriage Analysis, Journaling Session, Thought Record and the rest), are coaching-only: never put into this block, so an ordinary work session never sees them. Coaching sessions still read them, and older write-ups are covered too.
A new artifact (or a bumped version) shows to future sessions; sessions already spawned before the save are not retroactively updated — Claude Code cannot amend an in-flight system context, and quietly re-prompting would be confusing.
Choosing a context level (Full / Key Points / Summary / Micro)
Every saved artifact is automatically condensed into four versions, so the same profile can be fed into a session at four sizes: Full (the complete text), Key Points (the essentials as structured bullets), Summary (a 2–4 paragraph executive summary) and Micro (a one-or-two-sentence one-liner).
The three condensed versions are generated in the background right after a save, from the whole artifact (up to its 100,000-character limit), using a fast, low-cost model with the same condensing prompts as ContextDock (copied into Omniscio, so nothing is sent to ContextDock's cloud and no account is needed). Generation is cost-capped: its spend counts against the $0.25-a-day compaction allowance it shares with live session summaries (part of the $0.35 session pool), and stops for the day once that allowance is 60% spent ($0.15), so the summaries that keep a long session coherent keep their share. It never blocks anything, and skips work it doesn't need: artifacts too short to condense are left alone, a condensed version that didn't come out shorter is discarded, and levels are re-generated only when the text changes. The exception: versions made earlier from only the first 10,000 characters of a longer artifact are condensed again once from the whole artifact, by the catch-up that runs after launch, within the same daily cap; until then, and whenever that refresh fails, the older version stays. Each condensed version is saved as it is made, so a crash part-way keeps the versions already made and the next run makes only the missing ones.
Who picks the level? By default the AI picks — the budget-aware selector above chooses the richest level of each artifact that fits. You can override it: open any artifact and use the labelled Detail level control near the top, picking Auto or pinning Full / Key Points / Summary / Micro to force that version into every session. A shorter level not yet generated is greyed out, and hovering explains that condensed versions are created after you save and need an Anthropic API key; the level already saved stays selectable even with no text, so you can always change your choice. Switching to a shorter level previews it in the pane under a banner naming the level — e.g. "You're viewing the Summary version — a shorter rewrite, not your full profile." with a Show full button — so a summary is never mistaken for your full profile. If the saved level has no text the pane shows the full text instead, with a notice such as "Summary isn't ready yet, so the full text is shown." Your pin is saved per artifact and survives relaunch.
The Context Budget pane (in the AI Coaching sidebar) offers the same choice for every artifact at once, with each level's size, and a line under its budget bar counting what is shared, such as "1 write-up · 3 memories", in the user's language. A shorter level not yet written is dimmed and says so on hover, and clicking it does nothing, because the AI would ignore it and keep sending the full text; the app refuses to store such a pin. An older pin to a level with no text counts as Full in the row and the budget total — what is really shared — and clicking it once clears it back to automatic. Under each row's level bars sits Coaching only: highlighted while the write-up stays inside coaching, with no level bar, so the row never looks shared when it is not. Click it to keep that write-up out of your other sessions; click a level on a Coaching only row to share it again at that level, the same choice as the write-up's own Detail level strip. Write-ups that are always private (Daily Check-in, Weekly Review and the sensitive interviews) have their level bars greyed, because they can never be shared.
Session feedback
After an interview completes, a feedback dialog shows which interview you just finished (the prompt title) and a 1-to-5 star overall rating, with three optional aspect ratings below it — conversation quality, artifact quality, relevance — and a comment field of up to 2,000 characters. Submit is disabled until you choose an overall rating; everything else is optional. Dismissing without submitting preserves your draft (rating, comment, aspect scores) and restores it when the dialog reopens for the same session; drafts are kept in local storage (capped at 50), survive a restart, and are cleared on sign-out.
The dialog asks by itself only once: after your first finished coaching interview, about that interview, after a 2-second delay, and only while you are in AI Coaching; elsewhere it waits and appears when you come back (phone and web clients get the same refresh). A later interview, a Redo, or an interview finished while earlier ones are already on record never opens it automatically. "First" is read from your own saved progress, so no extra setting stores it. You can rate any interview from the completed card's overflow menu ("Rate"), which always opens the dialog — even when the automatic one was shown or dismissed, and even with automatic asking off — and counts as asked, so the automatic ask never pops for that session afterwards.
"Newly completed" always means newly completed for this client: the trigger takes its baseline from a progress read when it connects, and a push arriving before that baseline adopts its own reload as the baseline instead of diffing against an empty set. Without that, a phone or web client connecting to an existing history read every interview finished months earlier as brand new and popped the dialog about two seconds after loading. A failed progress read keeps the baseline already held, so a recoverable error cannot produce that burst either.
A "Don't ask again" checkbox turns off the ask-for-feedback preference (default on). When checked (on submit or close), the auto-trigger is suppressed until "Ask for feedback after your first session" is turned back on in AI Coaching settings, and installs that ticked the older checkbox are moved to this preference once, at startup.
Feedback is stored locally, one column per aspect rating; rating the same interview again replaces your earlier rating. A rating stays until its conversation is permanently deleted or your personal-development data is purged, which clears all feedback. It is not sent to any external service, and the dialog says so under its question: "Stays on your computer. Helps tune your coach."
Growth Timeline
The Dashboard shows a Growth Timeline strip when the user has at least one artifact, plotting each chronologically as a dot on a horizontal line, sized by version depth (more revisions, larger dot).
- Dot positioning — each artifact sits at a percentage based on when it was first created (the save time of its earliest live version) relative to the earliest and latest artifact, so revising a reflection never moves its dot to the revision date; the month axis, day span and date labels use the same time. When all artifacts share a timestamp, dots spread evenly.
- Version depth — dot diameter scales from 8px (version 1) to 20px (highest version), in the theme accent colour.
- Month axis — spanning multiple months, the first and last labels sit at the edges; a single month centers.
- Summary stats — artifact count, total versions, and day span (only when artifacts span multiple days).
The strip does not render at all when there are no artifacts — no empty-state chrome. It is display-only and screen readers get it as one labelled image.
Coaching Analytics
The Dashboard includes a read-only Analytics section between the recommendations and the list of interviews. A stats bar sits above the panel (artifact, version, bookmark and session counts, the most-used artifact, and a Copy stats button that copies a plain-text summary) with the Growth Timeline below it. The panel shows four figures:
- Sessions per Week — one bar per Monday-start week (local time) that had a coaching session, labelled with that Monday, first and last labelled. It covers the current week and the eleven before it, so the oldest bar always counts a whole week: the window starts on the local Monday eleven weeks before this week's Monday, not 84 days back from today. Sessions are read by start date and grouped by local calendar day; each bar is proportional to the busiest week.
- Artifacts by Category — a horizontal bar chart of documents per theme, largest first, using the same theme names as Progress by category (Foundation, Deep analysis, Work and so on, with More for your own interviews and anything outside the six themes). Each document is counted once, under the theme of the interview that produced it, with the same one-interview-per-document list and forgiving title match, so a title differing only in case or punctuation still lands in its interview's theme, and a document with no interview goes under More. The bars add up to your document total; the copied summary lists the same theme names.
- Avg Messages / Session — every session's messages added up and divided by the number of sessions, so a busy week counts for more than a quiet one. One decimal place, under the charts.
- Coaching usage value — everything the coaching feature has used: every session in the AI Coaching project (interviews, check-ins and free-form chats, including deleted ones, so deleting a session never lowers it), plus the background coaching AI calls (summaries, memory, profile, insights, goals, check-in prompts). Labelled usage value because on a plan most of it is covered by the plan rather than billed. Shown above zero. A second line, billed to your API key, shows the part actually billed (sessions and background calls on an API-key account or a metered vendor, classified by the same out-of-pocket rules the Stats page uses) and appears only above zero. One known gap: sessions removed by the retention purge keep only a project-less spend ledger row, so a purged coaching session no longer counts here.
Until there is at least one weekly entry or one categorized artifact, the panel shows a short invitation to start a coaching session instead of empty charts, and each part stays out of the page until its own data has loaded. The figures load in the background once the Dashboard's primary content settles, and only on the desktop app: a phone skips them, so there the section shows the Growth Timeline alone. They load again on every artifacts-changed push (an import, Delete all, a saved interview), so the stats bar, panel and most-used title follow the data without reopening the Dashboard.
Every number is written in the app's display language — counts as 1,234 in English and 1.234 in German, the average as 3.5 or 3,5. When no week had a session the summary leaves its average out and the panel shows 0.0. The copied summary is plain text, ready to paste into a chat, an email or any text field.
Importing existing work
If you've already done coaching work outside Omniscio (a standalone Claude session, a Google Doc, anywhere else), you can import it into a prompt without re-running the interview. On the Dashboard, any available (not yet completed) prompt card shows a ⋯ overflow menu on hover (desktop only) with an Import existing work option. The dialog lets you paste your previously completed content into a textarea, or load from file — pick a .md, .txt, .markdown or .json file, whose contents load into the textarea.
The textarea enforces the same 100,000-character cap as artifact editing. Content arriving from outside it — a loaded file or a fetched URL — is truncated at that cap, and either way you get a warning toast naming the limit plus a "(truncated)" indicator next to the counter. The warning is attached to the single place imported content lands, so a future intake path cannot ship silently.
An unfinished paste survives an accidental close or refresh: the dialog keeps it as a draft for that interview and restores it when the dialog opens again or you switch back to that interview — never when the interview list refreshes in the background, so a check-in landing while you type cannot reset the box. Emptying the box drops the saved draft.
The Dashboard header's Import opens the same dialog with no interview chosen: pick the interview from the list at the top, and Import stays off until you do. Text you paste or load before picking stays in the box, and nothing is saved as a draft until an interview is chosen, so a write-up you already imported never comes back next time.
Click Import (or press Ctrl+Enter) and the content is saved as the artifact for that prompt — the prompt immediately shows as completed, exactly as if you'd run the full interview. A confirmation names the interview ("Imported into “Life Story”.") with a View button that opens the saved artifact. The post-save pipeline fires normally: Key Points / Summary / Micro levels are generated in the background, and the Core Profile refreshes if you've crossed the 3-artifact threshold.
Opened from the Dashboard itself rather than a card, the dialog first asks which interview to import into, listing only interviews with no write-up yet; when every interview has one, the list becomes a note that there's nothing left to import into, and Import stays off. The Dashboard's Custom button likewise stays once every interview is finished or under way, so you can always start an interview on your own topic.
Import is desktop-only — the overflow menu is hidden on mobile, and the feature is blocked from mobile and web clients. There is no CLI endpoint.
Bulk import
For users with many pre-existing artifacts (migrating from a Google Docs or file-based workflow, say), the Dashboard header's Bulk Import button opens a dialog handling multiple files at once:
- Select files — a native multi-select picker (
.md,.txt,.markdown,.json). Each file is read through a per-file path-deny check, capped at a size limit per file and a row limit per batch. The dialog enforces the batch cap at intake on both the file picker and the URL paste box, and says what it left out, rather than letting the grid fill past it and lose the whole call at Import time. Over-cap URLs, and a link that fails to fetch (most often "Google is not connected"), go back to the paste box so a large paste is never lost: once its error row is removed, or the good rows are imported, the box holds it, so fixing the cause and pressing Fetch URLs again is the whole retry. - Auto-matching — each file is scored against available (uncompleted) prompts by matching its file name and content title against each prompt's title and artifact title, with high / medium / low confidence from score thresholds and the gap to the runner-up.
- Confirmation grid — a table of each file, its auto-matched prompt (editable via a dropdown) and a confidence badge. Reassign any file or skip it; duplicate detection prevents two files targeting the same prompt.
- Import — each matched file's content is saved as a new version, a single artifacts-changed push is emitted (not one per file), and the Core Profile refreshes once.
Bulk import is desktop-only too.
KMS vault mirror
When enabled, every artifact save (new version or update) is mirrored as a read-only markdown note in your default KMS vault, making coaching artifacts searchable, taggable and bookmarkable alongside regular vault notes without copying anything by hand.
The feature is off by default. To enable it, flip Settings → AI Coach → "Mirror artifacts to vault". Three conditions must ALL be true: the mirror setting is on; KMS is enabled; and a default vault is configured. If any gate fails the mirror is silently skipped — no error, no toast, no log noise.
Mirror writes are debounced (3 seconds) and fire-and-forget with an in-flight guard. A KMS failure never blocks the artifact save pipeline. Each note lands under the ai-coaching/ folder with a slug derived from the artifact title plus an 8-character ID suffix (e.g. ai-coaching/weekly-review-abc12345.md). The body carries an HTML comment header (<!-- AI Coaching artifact: Title, vN -->) followed by the artifact content. If the note already exists at that path it is updated in place; otherwise a new one is created.
Each artifact has its own debounce slot, so a batch import mirrors every artifact, not just the last saved. A firing timer snapshots the slot's title, content and version before it writes, so a save arriving mid-write replaces the slot's data without being lost: when the write finishes and the slot holds a newer version, the mirror re-arms and writes again. Every write, re-fires included, runs the same fire step and version check, so the mirror keeps re-arming until the vault note holds the latest saved version. A failed write is logged and heals on the next save. A timer still pending when the app quits is dropped — the database copy is canonical and the next save re-mirrors it. The existence check and the write are not atomic, but the per-artifact debounce serialises writes and the ID-suffixed path keeps different artifacts apart, so a duplicate create can only fail on the vault's unique-path rule.
The mirror's log lines name the artifact by ID only — never its title, slug or vault path — because a custom interview's title is the person's own private topic and main.log outlives a delete. A KMS error is logged with the slug replaced by artifact-<id>.
Related
The interviews that produce these artifacts, and the Dashboard you start them from, are on the AI Coaching sidebar (interview prompts and artifacts) page. What the coach remembers between them is in part 3, and the CLI surface plus the implementation pointers are in part 4.
Last verified 2026-10-20