First-time setup
Omniscio runs a setup flow the first time you launch it, walking you through a short numbered list of steps. Most people answer a few questions, connect a Claude account, and finish with a working session and a default project already created. If a step fails instead, two pages cover it. The app saying Claude Code isn't installed (or can't be found), a locked or upgraded database, or a machine that drags under too many sessions: Omniscio won't start, or says Claude Code is missing carries the fix for each, in the same plain language as this page, and names the one thing never to do — delete your database. If you are holding the exact words instead — This step hit a snag, Repair Needed, a sign-in refusal — Error messages explained quotes the common ones and says what each means and where to go next.
On this page
What it is
The first-time setup is the flow that takes a freshly installed Omniscio from "the app just opened for the first time" to "I have a working session running against my own code." It is Setup v2 (the team calls it the Cinematic Cascade) — the one first-run experience, a single calm card stepped through on a rail of named steps. Plus a few opinionated startup defaults, it makes this almost-zero-touch — most users answer a handful of questions, connect an account, and land in a working dashboard with a default project already created.
Setup is one card with a rail of named steps down its left side: Make it yours, About you, Your tools, Permissions, Connect your AI and Ready, plus Meet Tasks just before the end when the Tasks feature is turned on. That is up to seven screens: Permissions appears only on a Mac, Your tools only while its feature gate is on, and Meet Tasks only with Tasks on. The tools Omniscio needs install in the background from the first screen — a strip along the bottom of the card shows their progress — so there is no install screen to wait on.
Behind the scenes Omniscio also auto-creates the Claude project at ~/Claude so general-purpose chats have a home from minute one, and a default "Omniscio" sidebar group seeds the built-in virtual projects (Recipes, Skills, CLI Tools, Cron Jobs, etc.) so they're grouped instead of cluttering the top of your project list. When setup finishes it drops you in the Inbox, where two pinned welcome cards head the list above the starter agents it has already prepared.
Setup installs CLI tools only — it never installs, probes or mentions the Claude Code extensions. Both of those are now optional: Playwright MCP became a bundled, composed per-session built-in, and the Superpowers plugin was superseded by Omniscio's own built-in Dev Pipeline. The post-onboarding extensions banner still exists (it lists genuinely-required missing extensions) but is currently quiet, because no extension is required any more. Both are still listed under Settings → Extensions if you want to install one by hand. See mandatory-claude-extensions.md for the full story.
Where to find it
How to use it
Launch Omniscio for the first time. Setup opens automatically over an opaque cover, so the app never flashes its empty shell behind it, and it starts in your computer's own light or dark theme. The rail on the left names every step; finished steps stay clickable so you can go back, and Back sits beside Continue at the bottom. Finish later (top right, or Esc) asks before it leaves, then drops a reminder in your Inbox.
Make it yours. Pick your language and your appearance — Match my computer (the default), Light or Dark — and the card re-skins as you click. One line underneath says your settings stay on this computer and your AI chats stay between you and your AI provider.
About you. What you do (a role chip, or a few words), and how you heard about Omniscio (optional). Your answer is saved as you move on.
Your tools (when enabled). A starter set of tools picked for your role, each on a switch; below it, Add more lists every other tool A to Z, each with a line saying what it does — click one to add it. Only what you keep shows in the side nav afterwards.
Permissions (macOS only). "Let Omniscio keep you posted": Notifications, and Opening Terminal for sign-ins. Each shows whether it is allowed, and Allow brings up macOS's own prompt right then — for Terminal, so macOS doesn't stop a sign-in halfway (no Terminal window is left open). Both are optional; Continue always works. Voice, hotkeys and screen recording ask when you turn them on.
Connect your AI. Every AI is a chip — Claude (picked by default), ChatGPT, OpenRouter, Cursor, Gemini, Grok, DeepSeek, GLM — and each one is laid out the same way. An AI with a sign-in shows one Sign in with … button; for Claude it opens a Terminal window running Claude's own sign-in, and a short line says so once you click (on Linux it shows the one command to run,
claude auth login). Use my own API key opens one key field right there (the key is checked and stored encrypted on this computer). If the AI's own check rejects the key, the message stays on screen and Continue becomes Save anyway, which keeps the key without checking it again (changing the key brings the check back; a key in the wrong format, such as an OpenAI key in the Anthropic field, is never offered it), and a key that could not be saved says so instead of moving on; an AI that only takes a key, like OpenRouter, shows the field straight away. Start free, bottom left beside Back and Continue, moves on without either. If Claude is still installing when you sign in, the button reads Finishing install… and the sign-in opens by itself as soon as it is ready. An account already on this computer is detected, and you just continue. Keys for more AIs are added later in Settings → Accounts. See add-a-claude-account.md for the full account model.Nothing to configure before you go in. Setup does not ask how much your agents may do on their own — that is a decision for after you have seen one work. A new install starts on Guarded, the level that has agents ask before changing or deleting files, and you can pick Read-only, Autonomous or Full trust at any time under Settings → Workflow → Agent permission level; see agent-permission-level.md.
The required tools usually arrive before setup opens. On Windows, the Omniscio installer installs Claude Code and Git before the app first opens — using only routes that never ask for permission (Anthropic's own installer, and Git for Windows' per-user installer) — and skips any that are already there. It does this on a fresh, per-user install only (not on updates, not when installing for all users), gives up after about eight minutes rather than holding the installer, and never fails the install over it. Everywhere else (Mac, Linux, the portable Windows build), Claude Code starts installing the moment setup opens.
Meet Tasks (only when Tasks is turned on). If you have switched on Tasks in Settings → Lab, setup adds one short step that introduces it: capture anything in a second, plan your day, and find Tasks in the sidebar. Press Continue to finish setup as usual, or Finish and open Tasks to end setup right there and land in Tasks instead of the Inbox.
Ready. Your starter agents are waiting in your Inbox — reply to one to get it going — and the toolbar's feedback button is where to tell us what breaks. Go to Inbox finishes setup; it waits for your click, and while the tools are still installing it reads Finishing setup…. You land in the Inbox (or in Tasks, if you chose Finish and open Tasks), where two pinned welcome cards head the list above the starter agents.
Run it again any time from Settings → Setup Wizards → Re-run Wizard or the toolbar's Tours menu.
(After onboarding) The extensions banner stays quiet for now. Omniscio has a launch-time amber banner for missing required Claude Code extensions — but no extension is required any more (Playwright MCP became a bundled per-session built-in; Superpowers was superseded by the built-in Dev Pipeline), so you should not see it. If one ever does appear, click Install or Later to snooze it for 24 hours. See mandatory-claude-extensions.md.
Open Settings to fine-tune anything. The gear icon in the toolbar (or the Settings virtual project in the Omniscio sidebar group) opens the full settings modal. Common first-day adjustments: notification sounds and silence schedule (notifications-and-silence.md), keyboard shortcuts, the Default Projects Folder for Quick Create projects, the Allow API Keys to Run Sessions toggle, and the MemPalace persistent memory toggle. See settings-virtual-project.md for the alternate sidebar entry to Settings.
What if Google Auth is required at your install? Omniscio has an optional Google sign-in gate behind the requireGoogleAuth feature flag. The default is false — most users never see it. When enabled and Firebase is configured, Omniscio asks for a Google sign-in before setup opens; when enabled but Firebase is missing it falls through silently so users are never locked out of their own app.
How it behaves
Moving through setup
The step list down the left names every step ("Make it yours", "Connect your AI", …) instead of showing bare numbers, and a step you have already reached is ticked and clickable, so you can go back and change an answer. Steps you have not reached yet are listed but not clickable, so a click can never skip past a credential or consent step. The count above each screen ("Setup · 3 of 7") always matches the list, including on macOS, where one extra permissions step appears. The list is built from the flow's own step order, so when setup is shortened the sidebar shrinks with it.
On Which AI do you use? the ten choices are laid out in two columns, each with the tool's name on one line and a short description under it, so every option — including Something else — is on screen at once without scrolling. I'm new / not sure sits first and is the one selected to begin with; it sets you up with Claude, the default.
Your first-day welcome guide
Once setup is finished and your AI works, one "Welcome to Omniscio" chat in Ask Omniscio waits in your inbox. If it could not be made for you (it failed to start), a "Ready for your 5-minute interview?" card appears once your AI works; its Start my interview button creates the same chat, and the card is offered once. The chat opens on a short welcome card with Yes, let's go and Not right now; on Claude no AI runs until you answer, and on another engine one opening turn on your own account writes the card. Say yes and it asks a few easy questions about how you use AI today, then helps you add your first project, do a first real task, connect your tools, find tutorial videos and pick a mission — asking before it looks at your computer or changes a setting. It is offered once, only to a brand-new install (or someone who re-runs Setup and never had one). Full detail: first-day-welcome-session.md.
The starter agents — and their spend cap
When the welcome guide could not be created (a launch that failed), finishing setup offers you a free starter agent instead: one inbox card asks whether to start a guided mission, and only the one you pick is created, in Agent Missions. It runs on Omniscio's own company-funded model lane — not your Claude account and not your API key — and it is parked waiting on your first reply rather than spending on its own initiative.
That lane is capped, so a starter agent can never run away with company spend:
| Cap | Amount | What it means |
|---|---|---|
| Per starter session | $0.50 | A single starter agent stops once it has spent $0.50. |
| Per person, across all of them | $1.00 | The starter agents together stop at $1.00 for a given account — once, not per month. |
The per-session cap is the one that normally bites. A starter agent that hits a cap simply stops — nothing is billed beyond it, and nothing is charged to you at any point. So if one of your first agents goes quiet partway through a task, the budget is the usual reason.
The per-person figure is a one-time grant, and it belongs to your account rather than to this installation. It is tallied on Omniscio's servers against the account you signed in with, so clearing the app's local data or reinstalling does not hand the grant back. When it is used up the starter agents say so — "Your free starter credit is used up. Connect your Claude account to keep going." — and adding your own Claude account is what unblocks the work: your own account has no starter cap at all, because it is yours.
This affects only the pre-spawned onboarding agents. A session you start yourself, and any agent you launch normally, runs on your own account and is never touched by this cap.
The one switch that turns the free starter lane off. Everything above — the mission card, the Start button, the one free session a refused first message opens, and the company funding of a starter session — is governed by ONE switch, with two doors onto it:
- This machine: Settings → Lab → "Pre-spawn onboarding missions on the company DeepSeek
route" (the setting
onboardingPrespawnEnabled), or from the command linePATCH /settings/onboardingPrespawnEnabledwith{"value": false}. Off means the offer is never raised, nothing can be started on the free lane, and an agent already on it is not funded on its next turn. - The whole fleet, with no new release: the operator sets the server config doc
app_config/onboarding_fundingto{ "mode": "off" }. Installs pick that up within a few minutes and behave exactly as above. Setting it back to{ "mode": "company" }(or deleting the doc) turns the free lane on again — a missing or unreadable doc always means "on", so a broken read can never take the offer away from a new user.
The suggest-only mission prompt remains as the fallback, and the step-by-step missions are always startable from Agent Missions on whatever AI account you have.
Who pays for the starter agents
Normally Omniscio pays: the starter agents run on a company-funded lane, which is what the
caps above are bounding. The team can change that fleet-wide without a new release — the
starter agents can instead run on your own account, exactly like a session you start
yourself, with no company funding involved. If that switch is on and your own account is not
ready yet, the starter agents are simply not created; they are never quietly put back on
Omniscio's money, and the suggest-only mission prompt still offers them. The same fleet doc
also carries the value off, which switches the free starter lane off completely (see the
switch above). Your own install can also pin the payer choice itself through the advanced
setting onboardingPrespawnFundingMode (auto follows the fleet setting, company and
own-account pin it; it has no Settings toggle — see the settings reference).
An install that pins company or own-account deliberately ignores a fleet off, and its
Lab toggle remains the way to switch that install off.
How it works
Setup is mounted globally by src/renderer/src/features/onboarding/setup-v2/SetupV2Mount.tsx, which renders nothing at all on an idle, already-onboarded app — it shows an opaque first-run cover while a first-run is pending and then lazily loads src/renderer/src/features/onboarding/setup-v2/SetupV2Shell.tsx. The step order lives in src/renderer/src/features/onboarding/setup-v2/setup-v2-nav.ts (SETUP_V2_FRAME_IDS and buildSetupNavPlan, which drops permissions off macOS, your-tools while its gate is off and tasks-intro while Tasks is off), and the rail names and the "Setup · N of M" eyebrow derive from that one list. Each screen is a component in src/renderer/src/features/onboarding/setup-v2/frames/ (PreferencesFrame, AboutYouFrame, YourToolsFrame, PermissionsFrame, ConnectFrame, TasksIntroFrame, ReadyFrame). The background install is setup-v2-install-store.ts. A screen that throws degrades to a skippable recovery card rather than crashing the whole first-run. A step that crashes degrades to a recovery card that prints the frame id and app version (for example connect · v1.2.3, the same frame id the onboarding funnel records as onboardingFailure:<frameId>:frame-crashed) and carries a Get help action that exports the one-click diagnostic bundle, because the setup takeover covers the whole app and Settings → Diagnostics cannot be reached while a first run is stuck. Get help is also in the setup window's top-right chrome beside Finish later, so it is reachable from every step; it saves the bundle to your Downloads folder and tells you where it landed (nothing is sent anywhere).
Sign in with your Claude account records the passthrough pathway and launches claude auth login through IPC.CLAUDE_LAUNCH_SIGN_IN, then watches for the credential to land. Use my own API key records the apiKeyOnly pathway and saves the key to the isolated apiKeyOnlyAnthropicApiKey slot after a live check — never the managed account pool. The shared usability predicate src/shared/account-usability.ts is also what lets the main-process checkClaudeLogin probe treat an Omniscio-managed account as a satisfied "Claude Code Login" — in-app OAuth never writes the CLI's own ~/.claude/.credentials.json, but spawned CLIs authenticate from env credentials anyway.
Where you land at the end is decided by src/renderer/src/features/onboarding/setup-v2/post-onboarding-landing.ts: always the Inbox — unless the user pressed Finish and open Tasks, which lands on the Tasks virtual project with its own toast — with a completion toast chosen by whether any starter agents are waiting for you. The no-agents-yet toast points at the inbox and its button opens it — the Onboarding Hub is always empty now, so the old "create your first project from the sidebar" told every new user to go build something in a place with nothing in it. There is deliberately no project step in setup — you add projects afterwards from the + at the top of the projects sidebar, and the auto-created ~/Claude project is already there if you want to chat without setting one up. See add-a-project.md.
The auto-created ~/Claude project is seeded by ensureClaudeProject() in src/main/services/claude-project.ts, called from src/main/index.ts at app startup right after the database is ready. The default "Omniscio" sidebar group that holds the built-in virtual projects (Recipes, Skills, CLI Tools, Cron Jobs, Session Search, Automations, Quick Replies, AI Coaching, PRDStack) is seeded by a fresh-install migration — see projects-sidebar.md. The mandatory-extensions probe runs on first startup and on every subsequent startup (gated by settings.onboardingCompleted so it doesn't double-prompt during setup) — see mandatory-claude-extensions.md. The Google Auth gate sits in front of setup entirely and is documented in 2026-04-16-google-auth-gate-design.md.
Connecting — and what happens if you don't
Setup will not tell you that you are finished unless you are. The Connect your AI step is a required gate, and every way of leaving it with nothing connected now says so out loud:
- "Start free" exists only while the run-on-us gateway is switched on. With it off (the default) the option is not offered at all — it stored no credential and used to toast "You're all set", which is the one thing the flow must never say to someone who cannot run an agent. What remains is a real credential path: your own Claude login, your own API key, or a provider account/key.
- Bringing your own key is not the only paid way to start (T-27). The API-key step carries one line under its explanation: "Prefer not to use your own key? You can buy {app} credit after setup, in Settings → Plan & Usage." A new user who has no Anthropic key, and does not want one, is therefore told about credit at first contact, and the line names the destination they can actually reach once setup is done.
- "Sign in with your Claude account" opens the Claude CLI's own sign-in in a terminal. If Claude Code is not installed yet the launch waits for the install step, and the install step then watches for the login to actually land — Waiting for sign-in… → Signed in as your address (your plan), or a plain we couldn't see your Claude sign-in finish. The starter agents are only started once a login is genuinely confirmed.
- The deferred sign-in keeps its own controls on the install step. While it waits, the step says to finish in the Claude window that opened (naming that it may be behind Omniscio) and carries two buttons — Open the sign-in window again and I've signed in — check again — so a closed or lost window is never a dead end. Once it has given up, the same block offers Try again and Back to Connect, instead of pointing out of setup at a Settings page the person has never seen. The main button reads Continue without signing in while the sign-in is unfinished, and pressing it first asks — "You're not signed in to Claude yet, so your agents won't run until you do." with Sign in now / Continue anyway — because that click is the moment the choice is dropped.
- A free Claude account is never confirmed as connected. Omniscio treats a free plan as no credential, so the install step says the same thing the Connect step says — This is a free Claude account, and agents need a paid one. Add your own API key instead — and offers the way back to the Connect step. It never shows Signed in to Claude for an account the rest of the app will refuse.
- The whole screen is one sentence, one button and the two ways out. "Your AI work runs on your own Claude account." above Sign in with your Claude account, and under it one quiet row with Use my own key and Set up later as small text links. Nothing else is on the gate — no pricing line, no explainer paragraph, no badge, no hint. Nothing here hands a normal user a shell command: a Having trouble signing in? disclosure appears only when it can help — after a launch that opened nothing, or on a computer that cannot open a sign-in window at all — and then it leads with a plain sentence.
- The paid-plan requirement is stated before you commit, and again where it bites. The normal Connect gate still carries no pricing line — the requirement is on the two headline cards of Choose your AI, where the vendor is picked. When a sign-in lands on a free Claude account, the step says plainly that agents need a paid one, makes Use my own API key the main button, and offers Continue without an AI as the quiet way out; either exit leaves the pathway unable to claim a working credential, so the next step is never a screen saying you are ready.
- A different AI is one click away from the Connect step. On the Claude path a quiet Choose a different AI link in that same row takes you back to Which AI do you use?; picking another tool there and pressing Continue lands on that tool's own Connect screen. It writes nothing and skips nothing.
- The sign-in watch is patient, and it never turns down a sign-in that landed. If it has not seen the credential within about ten minutes you get We couldn't see your sign-in, with a Try again that re-opens the sign-in and a Continue that reads the credential one more time before it decides — so a sign-in that finished a moment too late is confirmed instead of denied. You never have to wait it out: while the sign-in window is open you can click I've signed in — check again, or simply switch back to the Omniscio window, and either one looks immediately.
- The account that signed in is named — Signed in as your address (Pro) — so you can see it was the account you meant, not a different one. If it is a free Claude account, the step says plainly that agents need a paid one, makes Use my own API key the main button, and offers Continue without an AI as the quiet exit.
- You can always leave. A quiet Set up later link sits beside Use my own key under the primary, and it warns, in the same words as every other exit, that your agents will not run until you connect. Nobody is trapped here and nobody leaves believing they are set up when they are not.
- On a computer where Omniscio cannot open a terminal for you (Linux), the sign-in panel opens by itself and shows the single command to run in your own terminal — always below Use my own key and Set up later, never above them. When you have signed in there, click I've signed in — check again, or simply switch back to the Omniscio window: either one looks for the saved Claude login and moves you on as soon as it is really there. If it is not, you get the same We couldn't see your sign-in panel as above.
- If a sign-in launch opens nothing at all, the step does not leave you on a dead button. A plain panel appears in place: We couldn't open the sign-in window on this computer. Choose Use my own key, or Set up later — setup finishes either way, with both links still exactly where they were.
- Not ready to connect yet? Choose "Set up later". A quiet Set up later link sits beside Use my own key — and, for a provider that needs a key, under the disabled Continue — so you are never stuck on this step. It warns that your agents won't run until you connect, then moves on; you can connect any time from Settings → Accounts, and the "Finish setting up Omniscio" card below reminds you.
- If you picked Gemini or Grok on a computer that does not have that tool yet, the step does not offer a sign-in it knows cannot work. It tells you the tool needs a one-time setup on this computer, says you can finish that later from the settings, and leads with its API-key field instead. Neither of those two can run here until the tool is present, so nothing on the screen claims otherwise.
- Skipping a Connect step that hit a snag is labelled Continue without connecting, not "skip this step" — it is the credential gate, not an ordinary frame.
- A failed Claude Code install is passable, and says plainly that your agents cannot run yet.
"Finished" means the AI you chose can actually run — not that some credential exists. The one
question every closing surface asks is answered by resolveChosenEngineRunPath in
src/main/services/provider-setup/chosen-engine-readiness.ts, and it is answered about your
engine: the app-wide default when you have set one that is not Claude, otherwise the provider you
picked on the Which AI do you use? step, otherwise the older "is there a credential anywhere"
truth. A ChatGPT pick with a failed Codex install is therefore not "set up" because you happen to
have an unrelated OpenRouter key on the machine, and it stops being "set up" if the Codex CLI
disappears later. A Free claude.ai plan is not a run path either — it signs you in but cannot run
Claude Code; Pro, Max, Team, Enterprise, an API key, and a plan we cannot read all still count.
The engine has to be runnable, so the same probe asks the app's own spawn gate
(getProviderSpawnReadiness) plus a real binary check — including the Claude Code binary for every
engine that runs it (DeepSeek, GLM, OpenRouter, GPT, xAI), because those vendors declare no binary
gate of their own. On Windows, Git is part of that gate too. All of it is derived in the main
process, so it survives a relaunch without the setup screen being open.
If setup finishes without a working way to run agents — nothing connected, or a core tool that would not install — the closing message says so (with a button straight to the pane that fixes it) instead of congratulating you, and the "Finish setting up Omniscio" card appears in your inbox. Its Finish setup button opens setup straight on the Connect your AI screen (or on the Install screen when a core tool is what is missing), never back at the language screen, so you do not click through steps you already did.
That card clears itself the moment you can actually run a session — either because your chosen engine came good, or because an agent has genuinely produced output in the last week (a session that was merely created, or that was prepared and never run, does not count, and neither does a run from months ago: old evidence cannot hold the reminder back from a user whose account or engine stopped working since). It is deliberately NOT cleared by the company-funded starter agents: those are pre-spawned for everyone, so counting them would clear the reminder for a user who had connected nothing.
When you finish setup and cannot run yet, the closing message says it plainly — "You need an AI subscription or API key to run agents" — and its button reopens setup on the step that fixes your case (Connect for a missing account, Install for a missing tool). The "Finish setting up Omniscio" card stays in your inbox behind it, and it comes back on the next launch for as long as you still cannot run anything, even if you archived it. That card's button always lands on the step that fixes the problem, not on the first screen. The one exception is a completion the main process could not answer at all: that is not the same fact as "you are set up", so it gets one neutral line telling you to check, and never "you're all set".
The card names the fix that can actually work. It is not always a missing account: if you are signed in, running on Omniscio credit, and that balance is empty, connecting an account would not unstick anything — so when this build offers credit top-ups the card says so and carries an Add credit link straight to Plan & Usage, with connecting your own Claude account as the alternative. Where top-ups are not offered, the card keeps its "connect an account" wording, because that really is the only road left. Both the card and the closing message read the same main-process answer, so the advice can never disagree with whether a session would actually start.
Where you land, and what is open. You always land in the Inbox with the welcome video open — not whichever first-run row happened to be written last. When your own AI account works, the closing message names your guide instead of describing the inbox: "You're all set. Your guide is waiting — open it and reply to start.", with a button that opens the "Welcome to Omniscio" chat. If you are part of a team, the same message keeps that instruction and adds one sentence saying you are also in your team's workspace.
If a message you send cannot start at all, Omniscio gives you the free road it already has. When a session refuses because there is no account connected and your Omniscio credit is empty, and the free onboarding lane is available, Omniscio starts a Free starter session for you on its own account — in the same project, carrying the message you just sent — and says so in the session you were in. It costs you nothing and draws on the same $1.00 free usage described above. Nothing is switched behind your back: the refused session keeps its own honest message and its Add credit / connect-account choices, and the free session is a separate, clearly named one. The offer is absent whenever it could not be honoured — you already have an account connected, the free lane is unavailable or funded on your own account instead, you have already started every starter mission, or one free session is already running.
Leaving and coming back
Setup remembers where you stopped. If you quit, walk away, or the app crashes part-way through, the next launch reopens the flow at the furthest step you had reached rather than at the beginning, and every step you already reached stays clickable on the rail — so you can move back and forth through what you already did. Answers you gave are kept independently, so a step you finished stays finished. Choosing Exit setup leaves your place where it was: the reminder card's button puts you back on that step, and finishing setup clears the remembered place.
What Exit setup promises, and when the reminder comes. The "Exit setup?" question says only what is true for everyone: you can finish anytime from Settings → Setup Wizards (its Re-run Wizard button opens these same setup screens again from the first one), and a "Finish setting up Omniscio" reminder waits in your inbox only if Omniscio cannot run your agents yet. If you can already run agents when you leave (a connected account, a working key, or the free starter lane), no reminder appears, because there is nothing left to nag about. Setup also does not reopen on its own at the next launch after you exit; the Settings route and the reminder card are the two ways back. The app stays fully usable after you exit.
Tasks sample tasks come only from setup. When Tasks is already switched on while first-time setup is still running (the setup then shows the Meet Tasks step), Tasks starts with four sample tasks. Turning Tasks on later from Settings → Lab, after you exited or finished setup, adds none, not even after a restart.
Two things are worth knowing when you come back:
- An already-finished Claude sign-in is noticed for you. Your Claude login happens in a separate window, so reopening the Connect step asks whether it is already there and shows you as connected rather than asking you to sign in again.
- The AI you picked is remembered with the place you stopped. If you chose a provider other than Claude and then stopped before connecting it, coming back resumes the Connect step on your provider rather than resetting you to Claude. Step back to Choose your AI if you want to re-pick it.
- The AI you pick becomes what new sessions start on. Choose ChatGPT, Gemini, Grok or any of the others and that becomes your default engine, so your first session runs on the thing you told us you use — instead of quietly running on Claude. Choosing Claude (or "I'm new / not sure", or "Something else") sets the default back to Claude. You can change it any time in Settings.
- If an engine we set up for you fails to install, we tell you. That CLI is a row on the Install step, so you watch it land and get a message naming what could not be installed plus a Try again that runs it again. Because the row is driven by the engine you picked — remembered across launches — quitting mid-install is not the end of it: when setup resumes, the step checks that CLI again and re-runs the install if it is still missing, exactly as it does for Claude Code.
A new computer knows you have been here before. Signing in on a second machine, or on one you just wiped, tells Omniscio that this account already finished setup somewhere. It then stops asking what it can already answer: the About you screen is skipped, the first-day welcome session and the tour are not offered again, the tutorial series does not restart from its first video, and the first-mission offer does not reappear. The first-day pause on Omniscio's own notices is also lifted, so a warning raised on day one of the new machine actually reaches you instead of being held back.
If the account has synced data or a cloud backup, one card appears in your Inbox offering to bring it onto this computer. The card opens Settings → Backup & Restore; nothing is restored, and no sync setting is switched on, until you choose it there. Your projects, history and settings live on the account that made them, so a fresh install starts empty only until you bring them across.
Related
- add-a-claude-account.md — what an account is (login vs API key), how to add more later, the active-account model
- add-a-project.md — the full Add Project dialog after onboarding (Quick Create, Browse Existing, GitHub clone)
- start-a-new-session.md — once you have a project, spawning and running sessions in it
- mandatory-claude-extensions.md — extension policy: Playwright MCP and Superpowers are both optional now, so onboarding installs neither, and the missing-extension banner stays quiet
- default-claude-project.md — the auto-created
~/Claudeproject that's there before you add anything - projects-sidebar.md — the default "Omniscio" group that holds the built-in virtual projects
- inbox-overview.md — once sessions are running, the Inbox is your daily triage view
Last verified 2026-11-12