Omniscio — Privacy Policy
Effective date: 2026-08-20 Last updated: 2026-11-16
Key terms at a glance
Plain-English summary — not a substitute for the full Privacy Policy. This box is a quick, friendly overview of the points people ask about most. It is not legal language and does not replace the numbered sections below, which are the Policy that actually applies. If anything here seems to differ from those sections, the full sections control.
- Local-first by design. Most of your data — your projects, agent sessions, the full text of your conversations, your settings, and any API keys you provide — stays on your own device and is never uploaded to us. The one exception is an own DeepSeek or DeepInfra key you save for cloud sessions, which is also kept encrypted on our server while it is saved and is withdrawn (made unreadable and no longer used) when you remove it — see §3. Separately, the built-in AI helpers you use send the text they work on to an AI provider, sometimes through our gateway — for example the daily digest and, if you turn it on, Tasks (§5). (§1, §2)
- Your keys stay yours. API keys and sign-in tokens are encrypted on your device. They are never sent to our servers — with one exception: an own DeepSeek or DeepInfra key you save is also stored encrypted on our server so your cloud sessions can use it without going through your computer, and it is withdrawn (it stops being readable by us and usable by your sessions) when you remove the key or turn the setting off. An Anthropic key you supply is not stored on our server at all. (§3)
- What we do handle in the cloud. A limited set: your account, billing and AI-credit metering, daily usage/spend aggregates, support messages, and — only if you turn them on — team and Shares features, and Cross-Device Sync (an end-to-end encrypted copy of your sessions, projects and settings that we hold but cannot read). Without those features, we do not upload your session history or your project code. (Three exceptions, all in §4: Cross-Device Sync, and two developer tools that are not part of the App you install.) The built-in AI helpers in §5 separately send the session text they work on to an AI provider. (§4)
- Diagnostics are always on. Basic reliability and usage telemetry (app version, device info, crash reports, feature counts, and the text of your Settings searches and "Ask Omniscio" help questions) runs with no off switch — this policy is how we keep it honest. The one exception is plugin feature counts, which have their own switch (on by default) in Settings. (§6)
- We don't sell your data or train on your code. We don't sell your personal information for money, and we never use your prompts, code, or conversations to train our own models. (The one "share": our public websites use Google Analytics and — when we advertise there — Meta and Google Ads conversion measurement, which you can opt out of.) (§4, §7, §8)
- Voice dictation is on by default, and so are recording captions. FlowVoice dictation ships turned on. It captures nothing until you hold the dictation shortcut (Ctrl + Windows key, or Control + Command on a Mac, unless you change it). While you hold it, your microphone audio is sent to the default Omniscio cloud speech engine and its providers to be turned into text, and that text — never the audio — goes through the default-on AI tidy-up. Turn FlowVoice off in Voice settings to stop this, or select the on-device Fluid Speech Mini engine to keep dictation audio on your computer. In-app voice chat and text-to-speech stay off until you turn them on yourself. Separately, the screen recorder writes captions automatically, on by default, and while the default Omniscio cloud speech engine is selected, the audio a recording captures — your microphone and, by default, your computer's system audio, which on a recorded call includes the other participants' voices — is streamed to Omniscio cloud speech and its providers while you record and again when the take finishes. Once a one-time notice has been shown, the finished transcript of each recording that has speech is also sent to OpenRouter to write its summary, unless you turn Auto-summarize recordings off. Turn off Captions for new recordings, or select the on-device Fluid Speech Mini engine in Voice settings, to keep recording audio on your computer. (§5)
- Personal-development features are sensitive and mostly stay local. If you use AI-coaching, journal, or Life Inventory, those notes and insights live on your device — unless you turn on Cross-Device Sync, which also carries your AI-coaching and Life Inventory records, end-to-end encrypted (§4). Coaching and journal reflections are generated by Anthropic — usually on your own Anthropic key, though a key you have already saved in the App, or a Company-held key on a plan that includes AI credits, can fund those calls instead. Life Inventory makes no AI call. The daily digest (on by default) leaves coaching and journal sessions out. (§2, §4, §5, §8)
- Your privacy rights. Depending on your state, you can access, correct, delete, or get a copy of your data and opt out of sharing — email legal@omniscio.com to exercise them. (§8)
- US-only for now; we store your data in the US — with one exception. The Services are offered only in the United States and the cloud data above is stored here, but shared video recordings are stored in Cloudflare R2's Eastern North America (ENAM) region — Cloudflare's placement for that bucket rather than a residency guarantee, so we do not promise they are held in the United States (§4). A few AI and voice vendors we route content to are outside the US too — §4 names them. Some of those you choose yourself, and some we use on our own keys for built-in features, as §4 explains. (§1, §4)
- How long we keep things. Local data stays until you delete it; cloud records like billing, diagnostics, and support have set retention windows. (§9)
- Google data you connect. If you connect your Google account, Omniscio uses that data only for features you can see, keeps it on your device, never uses it to train AI models, and follows Google's Limited Use rules. (§15)
1. Overview
Omniscio (the "App") is a desktop application published by Omniscio LLC ("the Company," "we," "us"), a North Carolina limited liability company. Omniscio helps you run and manage AI coding agents on your own computer.
Omniscio is local-first by design. Most of your data — your projects, agent sessions, the full text of your conversations, local cost tracking, settings, and any API keys you provide — is stored on your own device and is not uploaded to us, except an own DeepSeek or DeepInfra key you save for cloud sessions, which is also kept encrypted on our server while it is saved and withdrawn when you remove it (see §3). A limited, purpose-specific set of data is handled in the cloud to run accounts, billing, support, always-on diagnostics, the optional team features, and the optional (end-to-end encrypted) Cross-Device Sync. This policy explains both.
Where you are. The Services are currently offered only in the United States. This policy is written for US users; if we expand to other regions we will provide region-specific terms.
2. Data stored locally on your device
Stored in a local database (mission-control.db, in your OS application-data folder), a local config file (the credentials and connected-account identities in it are encrypted; the rest of the file is not), and data files in that folder, and not sent to the Company except as described in Sections 4–6:
- Sessions and conversations — your agent sessions and full message history (prompts, agent responses, system messages, attachments).
- Projects — names and local folder paths.
- Costs and usage (local) — per-session AI cost/token estimates and a local cost log (see Section 6 for what leaves the device).
- Automations, schedules, saved prompts.
- Channels/integrations you connect (e.g., SMS, Slack, Telegram, RSS, webhooks, email) — their message history and config.
- Notes/knowledge, settings, and credentials (Section 3).
- Personal-development data (only if you use those features) — the optional AI-coaching, journal/reflection, and self-assessment ("Life Inventory") features store the notes, ratings, and generated insight/profile text you create. This can include psychological, emotional, and health-adjacent information about you, and is kept locally on your device — unless you turn on Cross-Device Sync, which also uploads your AI-coaching and Life Inventory records, end-to-end encrypted (Section 4). The AI-coaching and journal-reflection features additionally send the relevant text to Anthropic to generate insights. Those calls normally run on your own Anthropic API key (the same key your coding agent uses), but a key you have already saved in the App, or a Company-held key on a plan that includes AI credits, can fund them instead, in which case we may receive that text. Life Inventory is computed on your device with no AI call. The daily digest (on by default) leaves AI-coaching and journal sessions out. (See Sections 5 and 8.)
- Dictation (FlowVoice — on by default, used only when you hold its shortcut) — your recent dictation history (the transcript and the app or window you dictated into) and the audio of each dictation, kept only until its text is saved. If a dictation could not be transcribed, or had to be recovered from its saved audio, the audio stays on your device so you can transcribe it again — until you delete it, it ages out with your data-retention setting, you sign out, or you erase all local data.
- Screen recordings (only if you use the recorder) — the recorder keeps each take's screen video, the microphone and system audio it captured, and webcam video when you record with the camera; the screenshots and "snips" you take; and the caption transcripts and AI summaries written for a recording. By default this is a
screen-recordingsfolder inside the App's application-data folder. You can move it to any folder you choose in the recorder's settings, and a recording you export is copied to the folder you pick. These are ordinary media files rather than database rows, so a copy you export or a store you relocate stays where you put it until you delete it there. - Local logs — diagnostic logs for troubleshooting. They stay on your device, except that error and diagnostic events from them are sent to us as described in Section 6, and a crash report or a report you send can carry a scrubbed excerpt of the log.
Some operational data also lives outside that folder. If you use the App's built-in developer
tooling (the dev pipeline / cloud test offload — see §4), it writes to a folder called .amc in
your user home directory: branch names, run history, job ledgers, and the App's local access
token. That folder is not inside the application-data folder, so removing one does not remove the
other.
To delete your local data: delete it in the App, remove the App's data folder, and remove the
.amc folder in your home directory if it exists. If you moved the screen recorder's folder to
another location, or exported recordings to another folder, delete those copies there too —
removing the App's data folder does not reach them. If a session ran in an isolated working copy, the
App kept a full copy of that project in a working-folders folder — beside the project, or in the
folder you chose for it — with its own dependencies; remove those folders too, because removing the
App's data folder does not reach them. One record inside that folder outlives an in-App
deletion or "erase all local data": the cost/spend ledger for money already spent, which is kept as a
billing record until the data folder itself is removed (Section 9).
3. Credentials and API keys (verified)
- API keys and OAuth tokens you provide are stored locally and encrypted at rest using your OS secure storage (Windows DPAPI / macOS Keychain / Linux libsecret), with an encryption marker on the stored value.
- Own AI keys you save. A DeepSeek or DeepInfra key you supply yourself is stored locally as above and kept, encrypted, on our server so your cloud sessions can use it without routing through your computer. This is controlled by a setting that is on by default and that you can turn off. Removing the key withdraws the stored copy, and turning the setting off withdraws every copy stored this way — in both cases the stored credential is disabled: it stops being readable by us or usable by your sessions. We never use these keys to train models, and they are used only to make the requests you ask for. An Anthropic key you supply is used on your device only and is never stored on our server.
- They are decrypted only in the App's main process to make the requests you direct; they are stripped before reaching the App's interface layer. Apart from the own-AI-key storage described in the bullet above, they are not uploaded to the Company's servers.
- BYOK keys are used solely to authenticate the requests you initiate to that provider and the requests of the built-in AI helpers you have switched on (for example the daily digest and Tasks), which try your own saved AI key first — so a helper that runs by itself, such as the background Today ranking, can use your key without you starting that request — and (when you ask the App to validate a key) are sent directly to that provider. The only Company-server copy is the encrypted own-AI-key copy described above, which is withdrawn when you remove the key.
4. Data we handle in the cloud
When you use features that need our backend (Google Firebase / Google Cloud), we process a limited set of data:
- Account and identity — user ID, email, display name, profile photo, plan/tier, and (for team features) organization and role. Your display name and photo usually come from the Google or GitHub sign-in you used, and you can change both in Settings. A photo that comes from that sign-in is just a link; if you upload a photo yourself, the image file (up to 5 MB) is stored in our cloud storage, where only you can read it directly, and the app hands it on through a private link to the places that are meant to show it (your own profile, and a team you belong to). A published video Share also shows it. While Show my name on shared videos is on — and it is on by default — every video Share's public page displays your account name and photo under the video's title, and a photo you uploaded is embedded in that page. Anyone who has the link can open that page, so treat it as published; turning the setting off and re-sharing or refreshing the link removes it. It is deleted when you delete your account.
- Billing and AI-credit metering — a billing/credit record (balance and holds), an append-only usage/metering ledger of AI requests routed through our gateway, a hashed key identifying your gateway access, and a Stripe customer identifier. Card/payment details are handled by Stripe, not stored by us.
- An own DeepSeek or DeepInfra key you save for cloud sessions — the encrypted copy described in §3, held in our encrypted gateway vault under your account and keyed to the provider it belongs to, so a cloud session can use it without routing through your computer. It is withdrawn (made unreadable and no longer served) when you remove the key or turn that setting off. We do not keep it for anything else, and we never use it to train models.
- Sign-in location and security notices — each time your account checks in, we resolve your IP address to a country and region using a geolocation database bundled with our backend (the address itself is not sent to a third party for this). We keep a current-location record for your account (the countries seen and up to 10 coarse browser/OS device labels, such as
chrome:windows), and we append a sign-in location history row on the first sign-in of each day (and whenever a new country or device appears) — country, region, whether the country was new, a per-account salted hash of your IP (not the address), and the time — that expires after about 90 days. We use this to email you a "new sign-in" security notice when your account is used from a new country or a new device, and to keep an internal security timeline of where the account has been used. The same country/region is used for tax and availability. You can object to this at any time: email legal@omniscio.com and we will switch location capture off for your account and delete the stored location history (§8, §9). - Identified usage and spend aggregates — for signed-in users, we upload daily aggregates of AI spend and usage (e.g., daily spend in USD, sessions started, active minutes, top features), keyed to your account. These are identified, not anonymous. (This corrects any impression that nothing derived from your local cost data leaves the device — the raw local cost log stays local, but these daily aggregates are uploaded.)
- Support and feedback — if you contact support or submit a bug report/feedback, we receive the content plus limited diagnostics; a single bug report may be copied to several processors listed in Section 5.
- Team/collaboration features (enabled by you) — Team Chat messages, reactions, and shared-artifact ("Shares") content and comments are stored in the cloud for real-time collaboration. Comment threads may be readable by other signed-in users of the Service. Every Share has a Who can open this link setting with two choices, and you can also add a password and cap the number of views:
- We mostly publish what you share as-is. For a screen recording, one limited automatic step runs before it is shared: its transcript captions, AI summary and chapter titles are scanned for secrets, and API keys, private keys, OAuth and API-key fields, bearer tokens, JWTs, connection strings and capability tokens are replaced with placeholders. Nothing else is scrubbed — the video frames and any text visible in them are untouched, and names, email addresses and file paths pass through exactly as written. Please check the artifact before you share it.
- "Anyone with the link" (the default). Anyone who holds the link can open the Share. Links get forwarded, and a short link can be guessed, so treat this as "anyone I give this to, and anyone they give it to", not as secret. Our Share pages ask search engines not to index or archive them, but a link anyone can open can still be copied or screenshotted.
- "Only my workspace". Only signed-in members of your workspace can open the Share. This is the closest thing to a private setting — there is no separate private, unlisted or public mode.
- Video Shares name their publisher. A video Share's page shows your account name and photo under its title while Show my name on shared videos is on (the default). Turn that setting off and re-share or refresh the link to remove them.
- If someone comments on a Share. Commenting on a Share requires signing in with Google. That creates a Google-linked sign-in record carrying the commenter's email address, display name and photo, and each comment is published on the thread with that display name and photo, visible to everyone who can open the thread. We also send a copy of the comment — display name, photo and a short text preview — to the Share owner's alerts and to any user you mention. The sign-in record has no fixed retention period: we hold it until the commenter asks us to delete it at legal@omniscio.com, and it is not linked to an Omniscio account unless the same email address already has one.
- Opening or playing a Share is recorded. When someone opens a Share we count the view, record when it was last opened, and — if the owner turned on notifications for that Share — tell the owner. A shared recording's public player also reports playback as it goes: on start, pause, end, and when the page is hidden or closed, with no notice shown on the page. Each play session stores a session id, the first and last time, the furthest point reached, seconds watched, the recording's duration, the viewer's country, and salted hashes of the viewer's IP address and browser user-agent (never the raw address or user-agent), and is deleted after about 180 days. A Share with a view cap also keeps one counted-viewer row per network, keyed by a salted hash of the network address, which expires after about 180 days. A browser that sends Do Not Track or Global Privacy Control stops the playback beacon: when either signal is present we store no watch session. The Share's owner sees both in the app's Share analytics.
- Retention for Shares is in Section 9. If something is sensitive, use Only my workspace (with a password) or do not share it.
- Team Chat AI sends channel text to an AI provider. Team Chat has AI features that read messages, and what they read goes to an AI provider (Section 5).
- An AI bot answering in a channel. If a channel's owner sets up a bot to answer there (a starter bot or a reply bot), the messages it answers — the triggering message, the surrounding channel messages and attachment file names — become that bot's prompt in a Claude session, so they go to Anthropic. Other members' display names are replaced with pseudonyms and messages posted by workspace guests are left out, but the message text itself and the file names are not removed.
- The channel summary and
@mcanswers. A channel summary is built only in a channel whose creator turned it on, and it sends our helper AI lane (OpenRouter, Section 5) that channel's newest 100 messages, one line each, with each author labelled Participant 1, 2, 3… instead of by name and email addresses, phone numbers and links removed — the message text itself is sent as written. An@mcanswer sends the question that was asked, plus the content of the shared boards every member of that channel can see. - If you use these features in a channel with other people, the channel text can be theirs — let them know before you do.
- Download and ad attribution — when you click a download link on our websites, we record which campaign or link brought you there (the
utm_*tags, the referring page and, from that, a channel such as "Google Ads" or "Facebook") and the first page of our websites you opened on that visit (its address only, such asomniscio.com/guides/…, with no query string) under a random one-time code that rides on the Windows installer link, plus a salted, truncated hash of your IP address used only for abuse rate limits. Only if you have advertising consent (see our Cookie Notice) do we also store the ad click ids (gclid,gbraid,wbraid,fbclid), Meta's_fbp/_fbcbrowser ids, a random browser id, your browser's user-agent and the page you downloaded from. On Windows, the App reads that code from the "downloaded from" note your browser attaches to the installer file in your Downloads folder (it reads only Omniscio installer files, and never reads Downloads on macOS or Linux), and at sign-in links it to your account — once, and only for an account created within 30 days after the click. The linked channel then replaces your self-reported "how did you hear about us?" answer in our internal revenue-by-channel reporting, and the linked first page lets the same internal report show which pages of our websites led to sign-ups and payments. If you have advertising consent, we may report that signup, and later purchases (with their US-dollar amount), to Meta through its Conversions API using a one-way hash of your account id and email; a record of each report is kept for 90 days. Withdrawing advertising consent removes the stored ad fields from your download clicks and cancels any report not yet sent; deleting your account deletes the link and those reports. - Inbound email and support storage — if you use agent-email or support chat, inbound email bodies and support/helpdesk conversations are stored server-side (see retention in Section 9). If a plugin you allowed to send email schedules a message for later, the message and its recipients' addresses are stored on our servers until it has been sent, then deleted automatically (see retention in Section 9).
- Cloud backup to Omniscio (only when you start one) — a portable backup you choose to upload is AES-256-GCM encrypted on your device (its key never leaves your computer) and stored in our Google Cloud Storage under your account, so we hold ciphertext we cannot read. It is kept until you replace or delete it; deleting it removes every stored version.
- Cross-Device Sync (only if you turn it on) — an off-by-default feature (Settings → Cross-Device Sync) that keeps your setup the same across your own computers. It uploads to an Omniscio-hosted cloud locker on Google Cloud Storage in the United States: your sessions and their full message transcripts (AI-coaching and journal conversations included), projects (including the local folder paths), automations and scheduled jobs, saved prompts, quick replies, bookmarks, tags, email-summarizer rules, RSS and webhook sources (their tokens are stripped), deploy profiles, attachments to those items, and your AI-coaching records (goals and check-ins, insights, conversation memory and summaries, bookmarks, artifacts, feedback and pinned context) and Life Inventory ratings and results. Journal entries and your synthesized coaching profile are not synced. Credentials and API keys are never synced. The payload is AES-256-GCM encrypted on your device before it is uploaded: the data key is wrapped to each enrolled device (and to a one-time recovery code you are shown once) and sealed by each device's OS secure storage, so the locker holds ciphertext we cannot read. A small device registry is stored in Firestore so your devices can find and approve each other: a per-device identifier, the label you choose for that device, its public key, and timestamps — this part is not encrypted, because it is how a new device proves which account it belongs to. Turning sync off stops further uploads; anything already in the locker stays until it is replaced or removed by a later sync, or until you delete your account (Section 9).
We do not upload your local session history or project code as part of using the App — unless you turn on Cross-Device Sync, which uploads an end-to-end encrypted copy as described above. Separately, the built-in AI helpers listed in Section 5 — such as the daily digest, Tasks if you turn it on, and Omni in Mission Control, which reads board items, comments, documents, sheets and answers people submit on a form — send the parts of a session they work on to an AI provider, and for Omni that includes other people's names and comments, sometimes through our gateway.
One exception, and it is not part of the App you install. The Company's own repository ships a developer tool ("cloud test offload") that our engineers can run to execute a build or test suite on Company machines instead of their own. When a developer explicitly turns it on, it uploads a bundle of the code in their working folder — including files they have not committed — plus the name and email recorded on each commit, to a Company-owned Google Cloud Storage bucket in the United States, where Company-owned worker machines unpack and run it. Bundles are deleted on a 7-day schedule and run logs on a 14-day one.
A second exception, of the same kind, is the master sync. The same source repository ships a
contributor tool ("master sync") that reconciles a developer's checkout with the team's shared branch.
It is off by default and, like cloud test offload, runs only from a source checkout — never from an
installation you download. When it cannot settle a conflict on its own it spawns paid coding-agent
sessions — a coordinator and workers — that read, edit and commit the conflicting files in the
contributor's checkout. By default those sessions run on the contributor's own default coding engine; the tool's own settings
(masterSyncProvider and the two model settings beside it) move the whole sync to another engine, and
a provider set there that the spawn refuses falls back to that default engine. The conflict payload
goes to whichever engine the sync runs on — if that is DeepSeek, our gateway sends each request to
DeepSeek itself first, which is China-hosted, and fails over to DeepInfra in the United
States. The request is metered on a Company-held gateway key when a contributor's developer model
subsidy is enrolled, and otherwise on the contributor's own key for that engine.
This is called out here for completeness and honesty, not because it affects you: both tools live only in the Company's source repository and are not included in the application you download and install, so no customer installation can reach either. The people whose code and identity they handle are the Company's own contributors, working on the Company's own code. (The rest of the DeepSeek footprint — the Company-key lanes, and the coding engines a customer chooses for themselves — is in Section 5. The built-in helper features never use DeepSeek.)
Where your data is processed (data residency). The cloud data described above is processed and stored in the United States. Our backend — Google Firebase / Firestore / Cloud Storage / Cloud Functions — runs in the United States on Google Cloud: Firestore and file storage are in us-east1 (South Carolina), and processing (Cloud Functions) and backups are in us-central1 (Iowa), with some backup storage in a US multi-region. Crash/error reports go to Sentry's US region. Shared video recordings are the one exception: when a Share carries a video (a screen recording, a capture project, or a feedback video), the mp4 is stored in Cloudflare R2 (the same provider already used for the agent-email relay — see Section 5), in Cloudflare R2's Eastern North America (ENAM) region. That region is Cloudflare's placement for the bucket rather than a contractual commitment, so we do not promise that a shared recording is stored in the United States. Mobile push notifications are delivered through Google Firebase Cloud Messaging (FCM), a global Google service, and, on Apple devices, Apple's push service, so notification content transfers to that provider's infrastructure. Content-free notifications are on by default, so a phone notification carries only a generic title and a reference to the item. If you turn them off, a paired phone's notifications can carry an SMS sender's name and phone number, a preview of the SMS text, and an email subject. If you access the App from outside the US, you understand your data is transferred to and processed in the US. Cloud test offload (the first of the two developer tools described in Section 4) is US-only too: its storage bucket and worker machines run in us-central1 (Iowa), so a contributor's code snapshot and the name/email recorded on their commits are processed in the United States. Master sync (the second) follows the engine it runs on: on a non-US engine such as DeepSeek a conflict payload can leave the United States, as Section 4 describes. Some AI and voice vendors are also outside the US, as the next paragraph explains.
Some AI vendors are outside the US. Being offered only in the US bounds where our customers are; it does not bound where a vendor is. A small number of the helper-AI and voice vendors in Section 5 process content outside the United States on Company-held keys — most notably CrofAI (China; a retired coding lane, existing sessions only), DeepSeek (China), Zhipu/GLM and Moonshot/Kimi / MiniMax (China), Fish Audio (China-founded; the default read-aloud relay), the Alibaba-operated model upstream some OpenRouter requests are served by, and SiliconFlow (China), one of the hosts OpenRouter can pick for the Gemma 4 model when you choose it for Instant Fix. The published Subprocessors page lists each one under "Where processing happens" with what it receives and when. None of them receives Google user data through the DeepSeek lane: that lane is switched off whenever a Google account is connected (see §15). We do not currently operate EU/UK regions or rely on Standard Contractual Clauses (SCCs) for those transfers.
Cookies (functional only). If you use the optional web-access feature (pairing a phone or browser to the App over your local network), the App's local server sets two strictly-necessary functional cookies. omniscio_web_token (HttpOnly, SameSite=Lax) keeps your paired browser signed in across page loads — its value is your session access token. omniscio_device_id (HttpOnly, SameSite=Lax, Max-Age of 400 days) is a persistent, server-assigned random device identifier: the App mints one for a browser that has not paired before, so it can tell your paired devices apart, show you which ones are paired, and let you sign an individual device out. Neither cookie is a tracking, analytics, advertising, or cross-site identifier, so both are exempt from cookie-consent requirements as strictly-necessary cookies (each is set as soon as you sign in to that feature, before any banner), and both are first-party — shared with no third party. Clearing omniscio_device_id through your browser's settings simply makes that browser look like a new device the next time you pair it. The App sets no advertising or third-party tracking cookies. Our public websites (omniscio.com and docs.omniscio.com) separately use Google Analytics to understand site traffic (and, when we advertise, the Meta and Google Ads conversion tags described in Section 4), loaded behind a consent banner — including these legal pages, whichever of our two sites you open them on. The consent gate itself loads on every page of both sites that carries analytics (the help library, the /welcome page and the account-email link page carry neither), but the Google Analytics script is fetched only once your choice permits analytics — so a visitor who has not accepted downloads no analytics script at all, and Google Analytics receives your IP address, user-agent and the page URL only from that point. No visitor is opted in by default: in every time zone, Google Consent Mode holds analytics and advertising storage denied until you accept, and no analytics or advertising cookie is set until you accept. See our Cookie Notice for details and opt-out options.
Website A/B tests (first-party). When we test two versions of a page on omniscio.com, you see one version for your browser session. Until you press Accept on the cookie banner, that is all: the version is kept in your browser's session storage (cleared when you close the tab), no identifier is created, and nothing about the test is sent to us. If you press Accept on the banner that names it, your browser's local storage then keeps which version you were shown (omniscio-exp) and a random anonymous visitor ID (amc-anon-id), so you see the same version each visit; our own server records, without cookies or any third party, that this anonymous ID saw that version and, if you click a download link, which version your browser was in (the download click does not carry the ID). If you then sign in to the App with Google or GitHub, the sign-in tab opens omniscio.com/welcome, which links the version you saw to your new account once, so we can tell which version led to sign-ups; that link is made only if you pressed Accept. We never collect your name or email through the website for this. You can withdraw at any time - press Decline on the omniscio.com banner, use the "Do Not Sell or Share" link, or turn on Global Privacy Control - and the visitor ID, the stored version and the first-view marker are deleted from your browser immediately, nothing further is recorded, and the opt-in has to be given again to resume; records we already hold can no longer be linked to your browser; the same happens if you change your choice in another tab. Your answer covers both sites — it is remembered in a cookie scoped to omniscio.com as well as in this browser's local storage — while the A/B version and ID stay in this browser on omniscio.com. A choice made on an earlier banner that did not mention the visitor ID does not count as consent to it; if you accepted analytics on that banner, the banner asks you once more about the visitor ID only, and your analytics choice is not changed by your answer. Anonymous first-view records are deleted after about 400 days.
5. Third parties and subprocessors
We use the following. Core services run the Services; optional and secondary ones receive data only when you use the feature they serve, except the few whose entry says they run by default.
Core subprocessors
| Provider | Purpose | Data involved |
|---|---|---|
| Anthropic (Claude) | Default AI provider powering agent sessions, and the fallback route for the screen recorder's automatic summaries. If you set up an AI bot that answers in a Team Chat channel, the messages it answers also become that session's prompt — including other members' messages. The optional personal-development features (AI-coaching, journal reflection) also call Anthropic to generate insights — normally on your own Anthropic API key (the same key the agent uses), though a key already stored in the App, including a Company-held key on a plan that includes AI credits, can fund those calls instead. Mission Control's What Next also asks Anthropic for Claude through our gateway, on your own Anthropic key or on a Company-held one — its answer, the lookups behind it, and its deeper look, which uses a more capable Claude model. | Your prompts, code, file contents, conversation context, and attachments — sent when you run an agent (directly or via our gateway) — plus, for the personal-development features, the coaching/journal text those insight calls process (on your own key, or on a stored / Company-held key when one funds the call). What Next sends the question, the board context, and the matching snippets from each source that person has connected and allowed — email, calendar, Team Chat, board items, GitHub, Jira, Linear, Asana, ClickUp, Notion, OneDrive, Dropbox, plugins, Vault notes, Google Docs and meetings. For a Team Chat AI bot, the triggering message, the surrounding channel messages and the attachment file names are sent; member display names are replaced with pseudonyms and guest messages are excluded, but the message text itself and the file names are not removed. |
| Google — Firebase / Google Cloud | Authentication, database (Firestore), cloud functions, hosting, our AI gateway (Cloud Run), secret storage — and, on the Gemini engine, the live speech-to-text for dictation — and bot protection (reCAPTCHA, through Firebase App Check) on the public pages of the Forms plugin | Account/identity, billing/metering, aggregates, support, diagnostics, and team data described above. For dictation, the live microphone audio of what you dictate (see the FlowVoice row). On a public form, the visitor's IP address and browser signals. |
| Stripe, Inc. | Payment processing and US sales-tax calculation | Payment method and billing details (held by Stripe); the Company receives a customer identifier and subscription status. |
| Sentry | Application crash/error reporting | Crash/error reports with scrubbed messages (see Section 6). |
| Resend | Transactional email delivery (bug reports, digests, notifications) and product activation / marketing email (the onboarding drip) | The email content the App sends (e.g., your bug report and diagnostics); for activation/marketing messages, your email address and the message content. |
| SMTP2GO | Transactional email sent by our self-hosted backend apps — account invitations and password resets | The recipient's email address, the message subject and body, and any attachment. That recipient can be an invitee you invited, who is not a user of the Services at all. |
| Twilio SendGrid | Transactional email for the Calendar Scheduler booking feature — booking confirmations, approval requests, reminders, and cancellation / reschedule notices | The invitee's email address, their name, and the booking details carried in the message, including anything they typed into the notes or the host's own questions. |
| Cloudflare | Agent-email relay (inbound/outbound) and its outbound send leg; transactional notices (account security and sign-in location, account deletion, tier changes, invitations) sent through Cloudflare's Email Sending API; storage of shared video recordings in Cloudflare R2; and bot protection (Turnstile) on public forms | Email routed through the agent-email feature (sender, recipient, subject, body and attachments); the recipient and content of the account notices above; the mp4 of every video Share — screen content plus microphone and system audio, and, when the camera overlay is used, the webcam video of you and anyone in frame composited into that same mp4 (held in Cloudflare R2's Eastern North America (ENAM) region — a placement Cloudflare applies to the bucket, not a residency guarantee); and, on a public form, your IP address and browser signals. |
| OpenRouter, Moonshot/Kimi, Zhipu/GLM, MiniMax, Meta (Muse), Qwen (via our AI gateway, on Company-held keys; OpenAI models are reached through OpenRouter) | Cost-efficient "helper" AI models for built-in features — inbound agent-email pre-screening/classification (OpenRouter, on google/gemini-2.5-flash-lite), session-title generation, session summaries and daily digests, screen-recording summaries and chapters, Plain Speak rewrites, contact-name inference, Team Chat AI (a channel summary built from up to that channel's newest 100 messages, and @mc answers from the boards a channel's members share — both carrying other channel members' message text), Supermail AI email filtering + "Catch me up" thread summaries, and Inbox Pilot (which reads the recent turns of an agent session to decide whether it needs your attention, and — when you have it drafting replies — writes the suggested reply), Omni and Mission Control AI helpers (form-answer triage, document and spreadsheet reading, item progress checks, advice, break-downs, board reports and the advisor quick question, which send board item names and columns including people columns, comments, teammates' names, document and spreadsheet text, and form answers from people who may not use Omniscio), and Tasks AI planning — ranking what to do right now, the daily check-in and its chat, Plan my day, the Break it down interview, the arrange pass, Plan chat, the session-end learnings distill, and the session-end catch scan (watching finished sessions for follow-up tasks) |
The prompt/message content those helper features process, routed through our gateway on Company-held keys. (OpenRouter is the helper provider; the OpenAI models it uses are served through it.) The built-in helper features never use DeepSeek, on any key, including a DeepSeek key of your own. Tasks AI planning asks for the OpenRouter helper lane but is not guaranteed to reach it: the App's key order decides which key pays and which vendor receives the request — an AI key you have saved, then Omniscio credits, then the Company allowance this row describes. It sends your task text, task-list context, About You planning profile, Tasks memory/learnings, check-in and Plan chat messages, and — from the session-end catch scan — up to about the last 12,000 characters of the session it reads (your turns and the assistant's). Plan my day also carries your day's calendar: meeting start and end times, plus event titles when Show meeting names in AI planning is on. This includes the daily digest, which is on by default and carries the first message of every session it summarises — except AI-coaching and journal sessions, which it leaves out. Team Chat channel AI is not guaranteed to reach this lane either — it follows the same key order, and it sends the channel transcript it summarises (up to that channel's newest 100 messages, each author labelled Participant N with email addresses, phone numbers and links removed) and, for an @mc answer, the asking member's question plus the shared board context. Zhipu/GLM runs on a Company z.ai key, and Moonshot/Kimi, MiniMax, Meta (Muse) and Qwen run through the pooled Company OpenRouter key, so those coding lanes are NOT limited to your own key — the "other AI providers you configure" bullet below describes the separate BYOK case. For screen-recording summaries, it receives the finished recording's transcript (up to about 12,000 characters, with timestamps), which can include the voices of other people on a recorded call transcribed by our speech providers. |
| DeepSeek (via our AI gateway, on Company-held keys) | Ask Omniscio help questions for people who have no AI account of their own connected, the starter agents offered during onboarding, and coding sessions you run on Omniscio credits — only while no Google account is connected. Connecting Google switches this lane off, and those features move to another provider. | What you type to those agents and the work they do on your computer. Never Google user data. |
| DeepSeek and DeepInfra (upstreams serving our DeepSeek lane) | The "DeepSeek" models above are served first-party by DeepSeek on a Company-held key, and DeepSeek is China-hosted — so a request on that lane leaves the United States. If one of them errors, our gateway automatically retries the same request against DeepInfra, also on a Company-held key and hosted in the United States. DeepSeek V4 Pro has no failover and stays on DeepSeek. Ask Omniscio's last fallback reaches a DeepSeek model through OpenRouter instead, under the same no-Google-account condition. | The same prompt/message content the DeepSeek lane processes. |
| CrofAI (retired coding lane) | CrofAI ran as a Pro coding lane on a Company-held key, receiving the whole session — every prompt, every file the agent read, and every tool result. It was retired on 2026-09-21; its endpoint no longer serves requests and no new session can select it, but it stays listed because existing CrofAI sessions and their recorded spend still resolve through it. CrofAI is China-hosted, so while such a session runs, that content is sent there. | The whole coding session's prompts, files and tool results, for a session that already uses the lane. |
| FlowVoice (dictation — Omniscio cloud speech by default, or Google Gemini, Omniscio hosted, or Fluid Speech Mini on your own computer) | OS-wide voice dictation (speech-to-text) into any app — on by default, and engaged only while you hold its shortcut — and the screen recorder's automatic captions, which are on by default and are transcribed through Omniscio cloud speech whenever that is the engine (the default), and otherwise on your own computer. Dictation uses the engine you choose in Settings. Omniscio cloud speech (the default) sends your audio through our own cloud to our speech-to-text provider, currently Soniox; if Soniox refuses the stream or fails before any words are transcribed, our cloud continues that dictation on our first backup provider, Inworld, then on Meta Muse Voice Transcribe if that one refuses too, and, only where it is configured, on Deepgram after that. Gemini transcribes in Google's cloud. Omniscio hosted (paid plans) sends it to our own server, which streams it to our speech-to-text provider for that engine, Deepgram. Fluid Speech Mini transcribes on your own computer, so the audio never leaves it. On Omniscio cloud speech and Omniscio hosted, an optional AI tidy-up (on by default, switchable in Settings) removes filler words and stray repeated words before the text is typed: it never rewords what you said, and when it is slow or unsure your words are typed exactly as spoken. | Live microphone audio of whatever you dictate, plus your dictation language (for example "en"). For screen-recorder captions, it receives the recording's audio — your microphone plus, by default, your computer's system audio, which on a recorded call or meeting includes other participants' voices. With Omniscio cloud speech, it goes from your device to our own cloud service, which streams it to Soniox — or, when Soniox refuses the stream or fails before any words are transcribed, to Inworld, then to Meta Muse Voice Transcribe if Inworld refuses too (and, only where it is configured, to Deepgram after that) so the dictation is not lost — and returns the transcript to you (the provider can change, and our Subprocessors page lists the current one). With Gemini, that audio goes from your device to Google's live speech-to-text API — on your own Google key, or, if you have none, through our relay on a Company-held key. With Omniscio hosted, it goes to our own server together with a coarse foreground app name (for example "Chrome"), and our server streams the audio to Deepgram and returns the transcript. With the AI tidy-up on (Omniscio cloud speech and Omniscio hosted only), the text of each dictation — never the audio — goes through our AI gateway to OpenRouter, which runs it on Cerebras with data collection refused; only the tidied text comes back. With Fluid Speech Mini, nothing leaves your computer. The window title is deliberately not sent on any engine: a title can carry a document or person name, an email subject, or a URL, so it is captured locally for your own dictation history only and stays on your device. |
| Supermail backend | Company self-hosted backend (US) for the Supermail mail client and its server-side contact enrichment — engaged only when you use Supermail. Connecting a mailbox copies that mailbox onto this backend so the app can search, thread and summarize it. | Your connected mailbox, mirrored server-side — senders and recipients including cc and bcc, subjects, full message bodies and attachments, sent mail, drafts and labels, plus the mailbox's stored connection credentials (encrypted) — together with the contact-enrichment lookups performed on the Company backend. Retention is in §9. |
| TypeSafe (via our AI gateway, on Company-held keys) | The primary lane for the "Jev" decision feature — a typed question-and-answer call with a confidence score rather than generated prose. It runs on its own when the features that use it are working (for example the email safety screen and the spam check on texts); an OpenRouter route on the same gateway is the backup when TypeSafe is unavailable | The situation text and the questions submitted to a decision call, which can include message and email text. |
| Omniscio backend apps (Company self-hosted) | Our own backend services for support chat (the helpdesk), the plugin marketplace, Mission Control project boards (including the Monday.com and Trello board connections that run through them), importing boards from tools you connect, and the relay that carries some Shares — each engaged when you use that feature | The support conversations, board items, members' names and email addresses, imported board data, marketplace requests and relayed Share content those features handle. |
| unpkg.com | Serves the React runtime into a viewer's browser for a Share whose document runs code — automatically, whenever such a Share is opened | The viewer's IP address and browser user-agent; no user content. |
| Hugging Face | Download of the on-device model files (embedding, captioning and on-device speech) the App runs on your machine | Your IP address and browser user-agent; no user content — only the model file is fetched. |
| GitHub (ffmpeg download) | The App ships no ffmpeg. The first time a recording needs transcoding and none is present, the App fetches one automatically, with no prompt: first through your package manager (winget Gyan.FFmpeg on Windows, Homebrew on macOS), then, if that fails, a pinned, checksum-verified build from github.com |
Your IP address and request metadata (to GitHub, and to the package sources behind winget and Homebrew); no recording or other user content. |
| jsDelivr | Serves files the Video Editing Suite plugin downloads to run on your computer — a Google Fonts typeface you pick, and the GPU runtime its on-device background removal needs — fetched only when the plugin needs one | Your IP address and browser user-agent; no user content. |
ffmpeg build server (ffmpeg.martin-riedl.de) |
The Video Editing Suite plugin fetches a pinned ffmpeg static build from this server on macOS and Linux the first time it needs one and none is installed | Your IP address and request metadata; no recording or other user content. |
Agent Client Protocol registry (cdn.agentclientprotocol.com) |
Download of the public list of coding agents you can add, when you open that list or add an agent (kept for a day) | Your IP address and request metadata; no user content. |
| LiveKit (real-time call media) — not yet active | In development; not enabled in the current build, so no data reaches LiveKit today. When Team Chat voice/video calls ship, LiveKit will act as the real-time audio/video relay (SFU), engaged only when you start or join a call, with the Company minting a scoped access token on its own LiveKit account (Company-operated, lane-2). LiveKit will be added to the Subprocessors list before that path is ever enabled. | Once enabled: live audio and video of the call participants (including other members on the call) plus call room/token metadata. |
Optional / secondary services
Most of these receive data only when you turn on or use the feature they serve; the few that run by default say so in their entry. Our Subprocessors page is the complete, current list.
AI sign-in / OAuth provider — authenticates your Claude/Anthropic account.
Other AI providers you configure (e.g., OpenAI/Codex, Google Gemini, DeepSeek, Moonshot/Kimi, Zhipu/GLM, MiniMax, xAI, OpenRouter, Qwen through Alibaba's DashScope service — hosted in Singapore by default — and other coding-agent engines you select) — receive your prompts/context when you use them, on your own API key, under their terms. DeepSeek cannot be used while a Google account is connected (see §15).
Channel integrations you connect — receive data only for the integrations you enable, each a direct connection to your own account. These include Pushbullet (SMS gateway — receives SMS/MMS content and synced contact info), messaging platforms (Slack, Telegram, and Discord, where an automation can post a message to a channel through a webhook you save), and meeting, content & automation tools (Granola, Fathom, Zoom, Loom, Canva, Vimeo, Zapier), plus the email/calendar accounts you link. Fireflies (on your own key, only if you turn it on) receives the search words and date ranges you ask about and returns your meeting transcripts, participants and summaries to the App. Composio (in development, off by default) connects your tools on your own Composio key; when your Google account is connected through Composio, your Google requests and the mail and file content they return pass through Composio. If you save a Vimeo access token, the screen recorder can upload a recording — its screen, audio and webcam video — to your own Vimeo account when you ask it to.
AgentMail on your own key — if you save your own AgentMail key (Settings → Email Inbound), the email features that still use it — Email Summarizer replies and the forward-email automation — send the message (recipient, subject and body) through your own AgentMail account. We hold no AgentMail account for your data.
SMS alerts you configure — if you set up Twilio in Settings so that off-screen alerts are texted to your phone, the alert's title and body (which can name a session or quote a message subject) and the From/To phone numbers are sent to
api.twilio.comon your own Twilio account. This is separate from the booking emails our Calendar Scheduler sends through Twilio SendGrid, which are sent on our account and listed under Core subprocessors.Board / project-management integrations you connect (e.g., Trello, Jira, Notion, Linear, ClickUp, Airtable, Asana, Monday.com) — when you sync a board, member names/emails and card content are exchanged with that service on your own account/key, under their terms. Monday.com data (including OAuth tokens, board items, and people-column names/emails) is additionally proxied through a Company-operated backend.
GitHub — if you sign in with GitHub, or connect a repository so the App can push a branch, open a pull request, or read PR/issue activity, that happens on your own GitHub account under GitHub's terms. Pushing a branch or opening a pull request transmits every commit's author and committer name and email address along with the code changes themselves, and GitHub sign-in shares your account identity with us.
- One GitHub read is on by default, and it does not wait for you to sign in or connect a repository. The PR-body check starts with the App and repeats every 10 minutes. Through the login your machine's GitHub CLI (
gh) already holds, it lists your open pull requests across every repository that login can see, by asking GitHub's own search API (search/issueswith the queryis:pr is:open author:@me, newest 30); for a pull request in a project you have registered in the App it then reads that pull request's body, its head branch, and its diff statistics. A finding is handed to the agent session that owns the branch — which sends that pull request's title and URL to the AI provider you have configured — or raised as an inbox card for you. Nothing here goes to us: the content moves between your device and your own GitHub account, and GitHub is the only remote party. It is on in development builds of the App and off in the release builds you download, and the Settings screen has no control for it. It can be turned off by setting theprBodyCheckEnabledsetting to false (through the settings API, the CLI, or an agent) or by theAMC_DISABLE_PR_BODY_CHECK=1environment variable.
- One GitHub read is on by default, and it does not wait for you to sign in or connect a repository. The PR-body check starts with the App and repeats every 10 minutes. Through the login your machine's GitHub CLI (
Cloud coding sessions (in development, off by default) — Agent37 on your own Agent37 key, and Box (ascii.dev) through your own Box login, run a coding session on a remote machine; starting one sends that machine the project's working tree, the agent session, and the credentials the session needs.
Cloud coding agents on your own key — Devin (Cognition) runs a session on Devin's own cloud and receives your prompts, agent context and code; Reason (Reason Machines) runs a session on its own cloud and receives your prompts, agent context and the name of your GitHub repository.
Grok Bot (xAI) — a desktop agent you sign in to on your own account. Its service runs on Anysphere's Cursor infrastructure and receives your prompts, attachment file paths and the agent's replies.
Archil cloud disks — if you connect your own Archil account, the App mounts an Archil disk for a session and runs commands and searches against it (
control.*.archil.com). The files you put on that disk are stored at Archil, in the region you choose (US by default; an EU region is offered).Ad Studio (plugin) — if you connect your own Google Ads or Meta ad account, the plugin reads and changes your campaigns through the Google Ads API and Meta's Marketing API on that account, so the campaign names, ad copy, budgets and targeting you work on go to that platform under its terms.
Design search you connect — Mobbin (design-inspiration search on your own Mobbin session) receives the search terms you type.
Investment prices (Yahoo Finance) — the Investments view looks up a price and history for each ticker you hold from a keyless public endpoint (
query1.finance.yahoo.com). No key or account identifier is sent, but the symbols are.Weather lookup (Open-Meteo) — a weather automation or the built-in Daily weather report sends the place name you type, and the coordinates it resolves to, to Open-Meteo's keyless public endpoints (EU-hosted).
Typefaces (Fontshare, Google Fonts) — the browser (web/mobile) surface, the report templates, the Decks plugin and deck exports load typefaces from these keyless font CDNs, which receive your IP address, browser user-agent and the page address; no account, key or content.
Cloud backup you connect (Dropbox, Microsoft OneDrive) — the backup archive is encrypted on your device before upload, so the provider sees ciphertext only; connecting the account separately transmits your account email address and display name to that provider in the clear.
Bank sync (the Coffer feature, via a SimpleFIN bridge) — if you connect a bank, a credentialed request plus your bank account balances and transaction history are exchanged with the SimpleFIN bridge you configure (self-hosted or a third-party-hosted bridge), on your own bridge subscription under its terms.
Contact/company enrichment (Supermail only) — our Supermail backend sends a one-way (MD5) hash of a contact's email address to Gravatar to look up a public profile image, and Supermail's contact pane loads a company's logo from Clearbit (
logo.clearbit.com), which receives the company's email domain and your IP address.Location search (OpenStreetMap / Nominatim) — if you use the card-location picker (in the project-management feature), your typed place-search queries are sent to the OpenStreetMap Nominatim geocoder (
nominatim.openstreetmap.org), a keyless public endpoint operated by the OpenStreetMap Foundation (EU-hosted). No API key or account identifier is sent.Automatic place detection (ipwho.is) — the first time you switch on the Daily Weather Report (or pin the header Weather widget), your public IP address is sent to the ipwho.is geolocation API (
ipwho.is), a keyless public endpoint, which returns a coarse location (country, region, city and postal code) so the App can fill in your place without you typing a ZIP code. No account, key, identifier or app content is sent with it, and the lookup is skipped entirely once you have typed a place yourself. The detected place is saved into your weather setting and you can edit or clear it at any time.SMS and Team Chat GIF search (Klipy) — if you insert a GIF in the SMS composer or in a Team Chat message, the search terms you type there are sent to Klipy, and a chosen GIF is fetched from Klipy's CDN; the search uses a Company-held default key unless you set your own. We have no data-processing agreement with Klipy, and Klipy's own privacy policy allows it to use what it receives for its own purposes, including advertising and creating new products — so treat a GIF search term as shared with that third party rather than held processor-only on our behalf.
Web-access and data tools (Firecrawl, Apify, X/Twitter) — when you use a built-in web-scraping, data-extraction, or read-only tweet/user-search feature, the URLs, queries, or page content you request are routed through our gateway on Company-held keys.
Bundled AI providers (OpenAI, Google Gemini, Perplexity) — an off-by-default "bundled API keys" feature would let your agent call these providers through our gateway on Company-held keys. All three are pre-integrated but not yet active — no data reaches them today — and they will be moved to the Core list before that path is enabled.
Image generation (JLS Image Studio — our own service) — an off-by-default, in-development feature sends your prompt and any reference images you attach, together with your account id, to our own image service (
jls-image-studio.web.app), which stores the images, galleries and share links it produces and forwards your request to whichever image-model vendor it runs for the model you pick (Google Gemini, ByteDance Seedream, Ideogram, OpenAI gpt-image or Meta Muse image today, with Topaz, Bria and 851-labs for upscaling and background removal). Prompts and reference photos are unredacted and can contain people or other personal data. The current model list is on the Subprocessors page. A separate Company-funded fal.ai image lane is declared in our gateway but is not used today — no request reaches it — and it will be added to the Core list before it is ever switched on.Unclaimed-property check (our own service) — an off-by-default feature sends the name, city and state you search to our own service, which queries the public state unclaimed-property registries. Any name you search for is disclosed to us and to those public registries.
Smrzz summaries (smrzz.com — our own service, hosted on Google Cloud / Firebase) — the built-in Smrzz library, on for everyone from 6 November 2026, sends nothing until you sign in to it. Once you do, it reads your smrzz.com library and sends what you choose to summarize to smrzz.com: the links you paste, the captions of a YouTube video read on your computer, a transcript or text you paste, a file you upload (audio, video or a document), or the video behind an Omniscio Shares link you paste, which is downloaded to your computer first and sent as a file. It also sends your deletes of summaries and, if you follow YouTube channels through Smrzz, the channels you follow and your email choices for them (checked every 15 minutes). If you ask for a video's speech to be read, smrzz.com fetches that video itself. To read captions and show pictures, your computer contacts YouTube (www.youtube.com and i.ytimg.com) directly with no sign-in, and fetches the picture smrzz.com names for a podcast or an article from wherever it is hosted. A pasted Shares link is opened on shares.omniscio.com or omnisc.io, and its video is downloaded from the Omniscio Shares storage (Google Cloud Storage or Cloudflare R2). Each of those hosts sees your IP address. If you turn on the Omniscio account for Smrzz, these calls go through our gateway on a Company-held key, smrzz.com also receives your Omniscio account id so your summaries stay in one library per account, and each summary is paid from your Omniscio credits at the price shown in the app.
Services smrzz.com uses to read what you summarize — to make a summary, smrzz.com sends the link you submit to Jina AI (to read a web page, including a Spotify or Apple Podcasts page) or to SupaData (to read a social-media post, or a link straight to an audio or video file), and the podcast episode id of a podcastindex.org episode link to PodcastIndex. These requests come from smrzz.com's servers, not your device, and carry no account details.
ContextDock — an off-by-default integration uploads the documents you store and the search queries you run in ContextDock (
contextdock.web.app), authenticated with a ContextDock-issued key; your account email is read back from that service to label the connection.Decks plugin picture and fact lookups — when the optional Decks plugin is installed and is building a slide, the subject text of that slide (which can name a real person, company or private project) is sent as a search query to Wikimedia Commons and Wikipedia, and to Openverse and Wikidata; cover art for a resolved subject may be fetched from AniList or Valve's Steam image CDN. These are keyless public endpoints — no account and no identifier beyond the request's IP address.
Docebo LMS — if you connect your own Docebo tenant, your OAuth token, administrator profile (including email address), course list and learner records are exchanged with
{your-subdomain}.docebosaas.comon your own account.Screen capture & "Snip to Text" — if you use the on-demand snip/OCR feature, the screen region you capture is sent as an image to Anthropic (Claude Vision, for text extraction) and, for an AI-generated snip title, to OpenRouter — only when you explicitly trigger it. An image cannot be automatically redacted, so anything visible in the region you capture (including other people's data) is sent as-is; avoid capturing sensitive on-screen content.
Meeting transcription — if you turn on meeting transcription, audio of the meeting — which can include other participants — is streamed to a speech-to-text provider (e.g., Deepgram) on your own provider key to produce a transcript (in shipped builds; otherwise the feature stays notes-only), and attendee names may be stored in your local notes. If you then run Enhance on that meeting, the whole transcript — including what the other participants said, with their names intact — is sent to an AI model provider to write the enhanced notes: by default OpenRouter running a GPT-5.6 Luna model on Company-held keys, with Anthropic as the fallback route. Enhance runs only when you ask for it (automatic enhancement is off by default). You are responsible for any notice or consent your jurisdiction requires from other participants on the call.
Screen recorder captions and summaries — the screen recorder writes captions for every new recording that captures audio, automatically and on by default (the Captions for new recordings switch). With Omniscio cloud speech selected in Voice settings — the default — the take's audio is streamed through our own cloud to the same speech providers dictation uses (Soniox, then Inworld, then Meta Muse Voice Transcribe if the earlier ones refuse, then Deepgram where it is configured) while you are recording, and the extracted audio is sent again when the take finishes; only the transcript comes back to your device. That audio is your microphone and, by default, your computer's system audio — so a recorded call or meeting sends the other participants' voices to those providers, and those people have no relationship with us. When you are signed out, or our cloud speech refuses the stream, the take is transcribed on your device with Whisper instead. Automatic summaries are also on by default. The first finished take with speech sends nothing: a one-time notice is shown first, and its own button turns the automatic pass off. After that, every finished recording with speech has its full transcript (up to about 12,000 characters, with timestamps) sent through our AI gateway to OpenRouter, which runs GPT-5.6 Luna on a Company-held key, with Anthropic as the fallback route, to write a short name, summary and chapter list — and that summary becomes the recording's title. Auto-title (off by default) separately sends the transcript to an AI provider to be given a short title when you turn it on. Turn off Auto-summarize recordings and Auto-title recordings in the recorder's settings to stop the transcript leaving; turn off Captions for new recordings (which stops all of it), or select the on-device Fluid Speech Mini engine in Voice settings, to keep recording audio on your computer.
Voice features (off by default) — if you turn on voice, speech-to-text streams your microphone audio to the provider you select (Deepgram, ElevenLabs, Groq, or Meta — whose "Muse Voice Transcribe" engine is also reachable through OpenRouter) on your own key; text-to-speech sends the response text to Fish Audio by default through a Company-funded relay, and to Soniox the same way if you select it (both are Company subprocessors for this path), or to xAI Grok / ElevenLabs / Speechify / Pika on your own key.
Voice conversation ("Layer 2") (off by default) — a Lab feature that lets you talk back and forth with an assistant that knows what is on your screen and can pull more context. If you turn it on, each spoken turn sends the names, status and summaries of your open sessions, the last ~4,000 characters of the active session's transcript, and any session-search snippets it looked up, to OpenRouter, which serves it on a Grok (xAI) model by default. That is the free text of your own sessions, so it can carry names, addresses, project details or other personal information — the App's default PII scrub does not rewrite free text. It uses your own OpenRouter key when you have one set and otherwise runs on a Company-held key. It is only ever started by you, and turning it off stops it.
In-app voice is OFF by default — FlowVoice dictation and recording captions are ON. The in-app speech-to-text and text-to-speech features are turned off out of the box. FlowVoice dictation ships on: it captures nothing until you hold its shortcut, and while you hold it your microphone audio goes to the FlowVoice providers in the table above; turn FlowVoice off in Voice settings to stop it. The screen recorder's captions also ship on, so with the default Omniscio cloud speech engine a recording's microphone and system audio goes to the providers listed above automatically, and once a one-time notice has been shown its transcript goes to OpenRouter for a summary (see Screen recorder captions and summaries). If you enable in-app voice, its audio and voice data are sent to the providers listed above. Dictation audio, in-app voice data and recording audio can all be sensitive personal information (see Section 8).
Other sites and services. The App, our websites, our emails, and content you view through the Services may contain links to third-party websites, tools, or services we do not operate. This Policy does not apply to those third parties, and we are not responsible for their content, practices, or privacy policies. Review the privacy policy of any third-party site or service before you provide it your information.
6. Diagnostics and product telemetry (always on)
Omniscio collects diagnostic and product-usage data to keep the App reliable. This collection is always on — there is no setting that turns it off, with exactly one exception: plugin feature counts, which have their own switch (see below). The disclosure here is how we keep it honest, not a promise you can disable it. Three settings control only the emails, never the collection: "Auto-Email Crash Reports" (Settings → Diagnostics → Health) controls the detailed crash email and the crash record sent with our usage telemetry — turning it off leaves crash capture by Sentry and every other telemetry leg running; "Email diagnostics to the team" (Settings → Diagnostics) silences the weekly usage digest and the crash email; and "Email crash reports" silences the crash email while keeping the weekly digest.
When it starts. This diagnostic collection begins when the App first launches — before, and independent of, any sign-in or Terms-acceptance step. Some fleet-health signals (for example, an install/first-run event, aggregate error/crash reports, and the diagnostic digests) can be sent before you sign in or accept these terms. Because the collection is mandatory, not consent-based, accepting the Terms and this Policy is how you are informed of it — it is not an on/off choice.
- What we collect: app version, operating system/platform, OS version, CPU core count, a coarse total-memory figure (rounded to whole gigabytes, never an exact size), whether the App's local database sits on a different disk from the operating system, feature-usage counts, crash/error reports, and the identified daily spend/usage aggregates described in Section 4. This also includes two typed-text streams — the text you type into Settings search and the questions you ask the in-app help / Ask-Omniscio assistant. Both pass through our scrubber first, which removes credentials and masks personal folder paths on a best-effort basis — it covers your home folder and the project and worktree folders the App knows about, but a path on another drive, or folders below the first one under your user folder, can survive — and they are keyed only to a per-install identifier (never your name, account, or a session). About that identifier: it is random, generated once per installation and stored on your device, and it is not derived from your name, email, or account — but deleting your account does not by itself delete it, and removing the App's local data (§2) resets it. A second, different identifier rides on AI requests that are routed through our gateway when you are signed in, so we can spot one machine abusing a free trial: a one-way fingerprint hash of the machine. It too is generated once per installation, stored on your device, and erased by the same reset. The Settings-search term is reduced to a short scrubbed snippet. The full text of your Ask-Omniscio help question is collected (credentials removed and personal folder paths masked on a best-effort basis, bounded to about 2,000 characters), so we can understand and fix the problems people are actually having — a help question can therefore include personal details you type into it, so treat it like a support message.
- Reduction and encryption: reports are automatically processed to reduce personal identifiers — we attempt to remove usernames, file paths, and message content, and error messages are truncated. This reduction is best-effort, not a guarantee that every identifier is removed (identifiers embedded in free-text error strings may remain). Our fleet-diagnostic payloads (sent to our diagnostics endpoint) are encrypted on your device before they leave it. Our crash-report and weekly-digest emails are sent over TLS (encrypted in transit) but their bodies are not additionally encrypted; in those emails the subject line carries only a shortened, hashed hostname token, while the full device hostname remains in the email body (so reports from different machines can be told apart).
- Plugin feature counts: Plugins may include anonymous counts of which features were used, with no content, account or repository details. These are the one part of the telemetry you can switch off: the "Plugin usage counts" setting (Settings → Diagnostics → Health) is on by default, and when you turn it off nothing is counted, nothing is queued, and any day already waiting is discarded — nothing is sent.
- What we don't do: we do not read your code or the content of your agent conversations for telemetry, and we do not sell this data.
7. How we use data
To provide, operate, and secure the Services; authenticate you; process payments and operate the AI-credit allowance; prevent abuse/fraud of credits; provide support; diagnose and improve reliability and features; and comply with legal obligations. We do not sell personal information and do not use your prompts, code, or conversations to train the Company's own models.
De-identified and aggregated data. Where we use aggregated or de-identified information, we maintain and use it only in de-identified form and do not attempt to re-identify it, except as permitted by law to test that the de-identification is effective. This paragraph does not describe the identified usage and spend aggregates covered in Section 4: those are uploaded keyed to your account and are personal information, not de-identified data, and they are handled as described in Sections 4, 6 and 9.
Disclosures for legal reasons and business transfers. We may disclose personal information when we reasonably believe it is necessary to: (a) comply with applicable law, regulation, legal process, or an enforceable governmental request; (b) enforce our terms and policies; (c) detect, prevent, or address fraud, security, or technical issues; or (d) protect the rights, property, or safety of the Company, our users, or the public as required or permitted by law. If the Company is involved in a merger, acquisition, financing, reorganization, bankruptcy, or sale of all or part of its assets, personal information may be transferred as part of that transaction; we will require the recipient to honor this Policy or will notify you of any material change to how your information is handled.
8. Your US state privacy rights
Depending on your state of residence, you may have privacy rights under a comprehensive state privacy law — including California (CCPA/CPRA), Virginia (VCDPA), Colorado (CPA), Connecticut (CTDPA), Utah (UCPA), and Texas (TDPSA), and other states as their laws take effect. Subject to the exceptions in each law, these rights generally include the right to know and access the personal information we hold about you, to correct it, to delete it, to obtain a portable copy, and to opt out of targeted advertising, the "sale" of personal information, or certain profiling. We do not sell your personal information for money. However, our public websites use Google Analytics with advertising features ("Google signals"), which discloses online identifiers to Google for cross-context behavioral advertising and is therefore treated as a "share" (and, in some states, a "sale") under the CPRA and similar laws. You can opt out at any time using the "Do Not Sell or Share My Personal Information" link in our website footer, by turning on Global Privacy Control (which we honor), or via the controls in our Cookie Notice. No other category of personal information is sold or shared, and none of the data inside the App is shared for advertising.
The California-specific disclosures below (the Notice at Collection and the sensitive-personal-information right to limit) are provided for California residents; the mechanics for exercising your rights, honoring opt-out preference signals, and appeals apply to residents of every state whose law provides them.
Sources of personal information. We collect the categories below from: you (what you type, upload, or connect); your use of the Services (automatically — feature usage, diagnostics, and device and network information); service providers acting on our behalf (for example, our payment processor); and third parties you connect or who contact you through an integration you enable (for example, the senders of messages you receive in a connected channel).
Categories of personal information (Notice at Collection):
| Category | What we collect | Purpose | Disclosed to service providers? | Sold or shared? |
|---|---|---|---|---|
| Identifiers | account ID, email, display name, profile photo (a link from your sign-in provider, or the image file itself if you upload one), IP address, device/install ID | account, login, security, support | Yes | Yes — website visitors only (IP address, Google Analytics / Meta browser ids and page activity on our public websites; see below) |
| Commercial information | plan/tier, billing and purchase records | billing, subscription | Yes (Stripe) | No |
| Internet/network activity | feature usage, diagnostics, crash reports | reliability, product improvement | Yes (e.g., Sentry) | Yes — website visitors only (IP address, Google Analytics / Meta browser ids and page activity on our public websites; see below) |
| Geolocation (coarse) | country and region inferred from your IP address on every sign-in, plus a coarse browser/OS device label. Kept as a per-account current-location record (countries seen and up to 10 device labels) and a sign-in location history that stores a per-account salted hash of the IP, not the address itself — see §4 and §9. | sign-in security (new-location and new-device notices) and an audit trail of where the account has been used; sales tax and availability | Yes — the region used for sales tax goes to Stripe Tax; the sign-in location history is not disclosed to a service provider | No |
| Payment/financial | payment-card details (handled by Stripe; not stored by us). Plus, if you use the Coffer personal-finance feature: structured financial data stored in your local encrypted database — bank balances, credit limits, a dated transaction ledger, and payee names (a SimpleFIN bank sync you connect is user-direct). | payment; personal budgeting | Card details: Stripe. Coffer data: local to your device; SimpleFIN sync user-direct | No |
| Sensitive personal information | FlowVoice dictation is on by default and captures audio only while you hold its shortcut; in-app voice and text-to-speech are used only if you enable them: microphone audio / voice recordings sent to your chosen speech provider, and response text sent to the text-to-speech provider. Also, by default, the screen recorder's captions send each recording's audio — your microphone and, by default, your computer's system audio, which on a recorded call includes other participants' voices — to those cloud speech providers. A recording made with the camera overlay also contains webcam video of you and anyone in frame; that video is not sent to a speech provider, and it leaves your device only if you share the recording or upload it to a service you connected. No SSN; your credentials stay on your device. | Provide default-on FlowVoice dictation when you hold its shortcut, the voice features you turn on, and the default-on recording captions | Yes — to the voice provider you select (and Fish Audio or Soniox, which read aloud through Company-funded relays), and, for recording captions, to the cloud speech providers named in §5 | No |
| Sensitive personal information (personal-development features) | If you use the optional AI-coaching, journal/reflection, or self-assessment ("Life Inventory") features: psychological, emotional, and health-adjacent notes, ratings, and generated profile/insight text you create. Stored locally on your device. AI-coaching and journal reflection also send the relevant text to Anthropic to generate insights — usually on your own API key, but a key you have already saved in the App, or a Company-held key on a plan that includes AI credits, can fund those calls instead; Life Inventory is computed locally with no AI call. The daily digest (on by default) leaves AI-coaching and journal sessions out. With Cross-Device Sync on (off by default), the AI-coaching and Life Inventory records are also uploaded, end-to-end encrypted (§4); journal entries are not. | Provide the personal-development features you use | Anthropic (your own, a stored, or a Company-held key) for coaching/journal insight generation | No |
Sharing for cross-context behavioral advertising. The online identifiers collected by our public websites (such as your IP address and a Google Analytics identifier) are shared with Google for cross-context behavioral advertising through Google Analytics' advertising features, and — when we run ads on those platforms — with Meta and Google Ads to measure which ads led to a download, signup or purchase (the Meta Pixel, the Google Ads conversion tag and our server-side reports to Meta described in Section 4). These are the only "sharing" we do. No visitor is opted in to this sharing by default: it starts only after you press Accept, in every time zone. You can opt out (see the "Do Not Sell or Share My Personal Information" link and Global Privacy Control described below, and our Cookie Notice). No other category in the table above is sold or shared, and nothing inside the App is shared for advertising.
- Sensitive personal information: we collect voice/biometric data only when you use a voice feature: FlowVoice dictation (on by default) while you hold its shortcut, the screen recorder's default-on captions while you record, and in-app voice features, which are off until you turn them on. You have the CPRA right to limit the use of sensitive personal information — exercise it by turning FlowVoice, recording captions and voice off in Settings (which stops any further voice data) or by emailing legal@omniscio.com. We use voice data solely to provide the voice features you use — never to infer characteristics about you, and never sold or shared.
- Personal-development features (AI-coaching, journal, Life Inventory): if you use these optional features, they can hold psychological, emotional, and health-adjacent information you record about yourself. This data is stored locally on your device — unless you turn on Cross-Device Sync, which also uploads the AI-coaching and Life Inventory records, end-to-end encrypted (Section 4). The AI-coaching and journal-reflection features additionally send the relevant text to Anthropic to generate insights — usually on your own Anthropic API key, though a key you have already saved in the App, or a Company-held key on a plan that includes AI credits, can fund those calls instead; Life Inventory is computed on your device with no AI call. You can delete this data in each feature's own screen, and the App's "erase all local data" function now removes these personal-development stores as well — including the on-disk AI-coaching files the feature writes under your
~/Claudefolder. Because those calls normally use your own API key, we do not normally receive this data — but we do when a stored or Company-held key funds the call. The daily digest (on by default) leaves AI-coaching and journal sessions out. We do not use this data to train the Company's own models. You have the CPRA right to limit its use — turn the feature off, delete the data, or email legal@omniscio.com. - Sign-in locations: each time you sign in, we resolve your IP address to a country and region and record it in a current-location record that also notes coarse browser/OS device labels, plus a short sign-in location history, to send you a "new sign-in" security notice and keep an audit trail of where your account has been used (see §4 and §9). The history is kept for about 90 days. You can object to this processing at any time — email legal@omniscio.com and we will turn location capture off for your account and delete the stored location history.
- Screen captures and meeting audio: the optional Snip-to-Text and meeting-transcription features can capture sensitive on-screen content or the voices of other people on a call. They run only when you enable or trigger them, and captured images are sent to an AI provider unredacted (an image cannot be scrubbed). The screen recorder is a third case, and it needs its own attention: its captions are on by default, so with the default Omniscio cloud speech engine a finished recording's audio — your microphone and, by default, your computer's system audio — is sent to our speech providers while you record and again when the take finishes; only when you are signed out or our cloud refuses the stream is it transcribed on your device (see §5, Screen recorder captions and summaries). The auto-title feature is in the same family: when you turn it on, that transcript is sent to our AI provider to be given a short title. Because a screen recording made with system audio captures the voices of the other people on a call, auto-title ships off by default and sends nothing until you enable it. Automatic summaries do not. They ship on by default: once the one-time notice described in §5 has been shown, every finished recording with speech has its full transcript sent to OpenRouter running GPT-5.6 Luna (with Anthropic as fallback) to write a short name, summary and chapters — and the summary writes the recording's title too, so turning auto-title off does not keep recordings AI-free. It is an opt-out after notice, not an opt-in: turn off Auto-summarize recordings to stop it.
- How to exercise your rights: email legal@omniscio.com (or use any in-App request tool we provide). We will confirm receipt and verify your request by matching it to the information associated with your account; we may ask for additional information to confirm your identity, which we use only to process the request. We will respond within the time your state's law requires — for California, within 45 days, extendable once by another 45 days when reasonably necessary, with notice. We will not discriminate against you for exercising these rights.
- Authorized agents: you may use an authorized agent to submit a request on your behalf. We will require written proof of the agent's authorization and may still verify your identity directly before acting.
- Right to appeal: if we decline your request and your state's law provides an appeal (for example, Virginia, Colorado, or Connecticut), you may appeal by replying to our decision or emailing legal@omniscio.com with "Appeal" in the subject line. We will respond with our decision and the reasons for it within the period your state's law requires (generally 45–60 days). If we deny your appeal, your state's law may allow you to submit a complaint to your state Attorney General.
Right to opt out of "sharing"; Do Not Sell or Share My Personal Information. Because our public websites use Google Analytics advertising features and Meta / Google Ads conversion measurement (a "share" for cross-context behavioral advertising, as described at the top of this Section), you have the right to opt out. To do so, use the "Do Not Sell or Share My Personal Information" link in our website footer, turn on Global Privacy Control, or use the controls in our Cookie Notice. We do not sell your personal information for money.
Opt-out preference signals (Global Privacy Control). Some browsers and extensions can send an opt-out preference signal such as Global Privacy Control (GPC). We honor GPC as a valid request to opt out of the sharing described above — when we detect a GPC signal on our website, we disable the Google Analytics advertising features, load no Meta or Google Ads tag, store no ad click id, and send no report to Meta for that browser.
"Do Not Track." Some browsers offer a "Do Not Track" (DNT) setting. There is no common industry standard for responding to DNT signals, so we do not currently respond to DNT signals; we honor Global Privacy Control instead (above). The App itself sets no advertising or cross-site tracking cookies (see Section 4); the advertising technologies on our websites are described in our Cookie Notice.
California "Shine the Light." California Civil Code § 1798.83 lets California residents request information about personal information disclosed to third parties for those third parties' own direct-marketing use. We do not disclose personal information to third parties for their own direct marketing, so there is nothing to report; you may still contact legal@omniscio.com with questions.
Nevada. Nevada law lets residents submit a request not to sell certain covered information. We do not sell such information, but you may direct a verified request to legal@omniscio.com.
Texas. We do not sell sensitive or biometric personal data for monetary consideration, and Texas residents may exercise the access, correction, deletion, portability, opt-out, and appeal rights described in this Section.
9. Data retention
| Data | Retention |
|---|---|
| Local data (sessions, messages, keys) | On your device until you delete it. Removing the App does not delete it by default — on Windows the uninstaller asks and defaults to keeping it; on macOS, dragging the App to the Trash leaves the data folder. Remove the data folder (and any session working-copy folders) yourself to erase it (see §2). |
| Personal-development data (AI-coaching profile/insights, journal entries, Life-Inventory results) | On your device until you delete it in the feature or run "erase all local data" — not automatically expired (see §8) |
| Account and billing records | Life of your account, then up to 7 years (tax/accounting/legal) |
| Sign-in location history (country/region, new-country flag, salted IP hash) | About 90 days, then deleted automatically. The account's current-location record is kept while the account exists and deleted with it. Opt out at any time (§8) |
| Cloud backup to Omniscio (encrypted on your device) | Until you replace it with a newer backup or delete it; deleting removes every stored version |
| Cross-Device Sync locker (end-to-end encrypted) | While your sync locker exists — sync keeps the latest version of each synced record (deletions included) and periodically compacts and garbage-collects older segments; deleting your account deletes the whole locker |
| Raw diagnostic events we receive in our fleet store | About 180 days |
| Installation-level fleet rollups (per-install daily spend/usage, keyed to the random install identifier rather than your account) | About 1 year for the per-install daily digests; the install record itself about 180 days after the install last reports; the raw events behind them are deleted at 180 days |
| Identified usage/spend aggregates | About 180 days, keyed to your account |
| On-device diagnostic logs | Rotated on disk — the 18 most recent log files: about a week of history on a busy machine, longer when the App writes less, and nothing older than 30 days |
| Support / helpdesk conversations | About 12 months |
| Inbound agent-email content | About 90 days |
| Scheduled agent-email messages | Until sent — the message body and subject are deleted at send, and the row with its recipient addresses about 5 days later. A message that is never sent is deleted about 33 days after its scheduled time. A send that keeps failing is set aside with its identifiers (no content) for about 90 days |
| Supermail's server-side copy of a connected mailbox (messages, bodies, threads, labels, attachments) | While the mailbox is connected to Supermail. Disconnecting a mailbox in Supermail deletes that mailbox's copy — its messages, threads, labels and attachments, and the contact details no other mailbox still uses. Supermail cannot disconnect your only remaining mailbox, and it has no in-app way to delete your Supermail account: to delete the account and everything stored for it, email legal@omniscio.com |
| Team Chat message history | No automatic deletion today — nothing deletes a Team Chat message by age, on any plan. A message stays until you or a channel admin deletes it, or until your account is deleted. Plan-based retention limits may be introduced when paid plans launch; if we do that, we will update this policy before any limit takes effect |
| Team Chat audit events (workspace settings changes, retention-policy changes and other admin actions recorded in the audit log) | About 730 days, then deleted automatically |
| Microsoft Teams history imported into Team Chat | Only messages from the last 365 days are imported; once imported they follow the Team Chat message-history retention above |
| Search index | About 13 months |
| Shared artifacts ("Shares") and their comments | Until you delete the Share — or, if you gave it an expiry date, until that date passes: from then on its link answers "expired" and no longer opens, its stored page, image and file objects are removed by our cleanup (about 15 minutes, while the App is running), and a video's files are deleted by a server sweep that runs every 6 hours, so within about 7 hours. A Share with no expiry date is kept until you delete it; we do not remove Shares by age. A Share opened with Anyone with the link may have been indexed or cached by search engines, and we cannot delete those copies |
| A shared recording's watch analytics (one row per play session: session id, first and last time, furthest point reached, seconds watched, duration, country, salted IP and browser user-agent hashes) | About 180 days, then deleted automatically. A viewer's Do Not Track or Global Privacy Control signal stops the watch session being stored |
| A view-capped Share's counted-viewer row (salted hash of the viewer's network address and the time it was counted) | About 180 days |
| Share commenter sign-in record (Google email, display name and photo, created when someone signs in to comment) | No fixed period — held until the commenter asks us to delete it (email legal@omniscio.com). It is not linked to an Omniscio account unless the same email address already has one |
| Download-click attribution records (campaign tags, referrer, the first page of our websites you visited, the one-time download code, a salted truncated IP hash, and — only with advertising consent — the ad click ids, Meta browser ids, browser id, user-agent and page URL) | About 180 days. The channel and first page we linked to your account at sign-up stay on your account until you delete it. Withdrawing advertising consent removes the advertising fields at once |
| Local cost/spend ledger (session id, account id, provider, amount, timestamps — no message content) | Kept by design after you delete a session or run "erase all local data", because it is the billing record for spend already incurred. It is removed with the App's data folder (§2) |
| Repository bundles and run logs from the internal developer offload (§4) | 7 days (bundles) / 14 days (run logs), by automatic schedule |
10. Security
We use technical and organizational safeguards: encryption of credentials at rest on your device (OS secure storage), encryption in transit (HTTPS/TLS), encryption of diagnostic payloads before egress, and access controls on our backend. No system is perfectly secure.
At-rest posture for your local data. Your local database (mission-control.db) and any saved user attachments are not application-encrypted by default; they rely on your operating system's full-disk encryption (e.g., BitLocker / FileVault / LUKS) as the compensating control, together with OS file-system permissions. An optional database-encryption feature can be enabled for defense-in-depth. (Credentials and API keys are separately encrypted at rest as described above.)
Breach notification. If a breach affects your personal information, we will notify you and the appropriate authorities without undue delay and within any timeframe required by applicable law. We maintain an internal incident-response process (detect → contain → assess scope → notify affected users and regulators as required → remediate → document). (We deliberately do not commit to a single fixed number of days here, because US state breach-notification deadlines vary; "as required by law" covers the strictest applicable one.)
11. Data-processing addendum (business/team customers)
If you use Omniscio on behalf of an organization and we process personal data on your behalf through the team features, a data-processing addendum (DPA) governs that processing and lists our subprocessors. A DPA is available on request for business/organization customers — contact us at the address in Section 13. We are putting data-processing agreements in place with the subprocessors listed in Section 5: agreements are in effect with our core platform processors (Google Cloud / Firebase, Sentry, Resend, and Cloudflare), and we are securing or confirming them for the remaining subprocessors. We maintain an internal data-processing register that records the current DPA status for each subprocessor, and we publish the current list on our Subprocessors page.
12. Children's privacy
The Services are not directed to children under 18, and we do not knowingly collect their data.
13. Contact
Privacy questions or requests: legal@omniscio.com, Omniscio LLC, 5301 Terminal St, Charlotte, NC 28208.
14. Changes
We may update this policy; material changes are indicated by the "Last updated" date and, where appropriate, additional notice. Continued use after changes take effect is acceptance.
15. Google user data
This section explains how Omniscio handles data it receives from Google when you choose Connect Google in the App. It applies in addition to the rest of this policy.
What we access, and why. Omniscio uses Google data only to provide features you can see and use in the App:
- Gmail — to show your inbox in Omniscio and let you read, search, reply to, forward, draft and send email; mark mail read or unread; archive, label, star, snooze, or move threads to trash or spam; and clean up your inbox by archiving or unsubscribing from senders. When you ask, Omniscio creates a Gmail filter that auto-archives future mail from a sender. Omniscio never permanently deletes mail from your mailbox — removing a message moves it to Trash, and permanent deletion only happens when you do it yourself in Gmail (or another mail client), never through Omniscio. (A draft you are writing and delete in Omniscio is removed from Gmail's Drafts.)
- Google Calendar — to show your calendars and let you, or an agent you direct, create, edit, move and delete events, add Google Meet links, find free time, and see your daily agenda.
- Google Drive, limited to files Omniscio creates or that you open with it — to publish documents and spreadsheets from Omniscio and keep them up to date.
- Google Drive folder names (read-only metadata) — only if you choose Drive folders for Omniscio to watch for spreadsheets, to list the files in those folders. It reads file and folder names and details, never file contents.
- Google Docs and Google Sheets — to open, import, create and edit the documents and spreadsheets you choose.
- Google Contacts (read-only) — to suggest recipients while you write an email.
- Google Meet (read-only) — only if you turn on the Google Meet feature, to list your meetings, participants, recordings and transcripts in Omniscio.
Where it's stored. Google data is processed on your own computer. Any local copies, such as a copy of your mailbox for fast search, stay on your device, and contacts are held in memory only. Your Google access token is stored on your device, encrypted with your operating system's secure storage, and is never sent to Omniscio's servers. The one exception is the Calendar Scheduler plugin: when you connect Google Calendar to it, a Google refresh token is sent to and stored on the Calendar Scheduler server, encrypted, so it can check when you are free and add booked meetings to your calendar while your computer is off; it only asks Google for your name, email address, free/busy times and the events on calendars you own, and choosing Disconnect revokes that access at Google and deletes the token. If you turn on cloud backup or cross-device sync, that data is end-to-end encrypted on your device before it is uploaded, so Omniscio cannot read it.
When it leaves your computer.
- AI features. When you use an AI feature on your Google data — for example a thread summary, reply suggestions, Email Cleanup suggestions, the daily digest, or asking an agent to read an email or check your calendar, or Plan my day in Tasks fitting your tasks around today's calendar events (it sends their start and end times, and their names only if you turn on Show meeting names in AI planning) — the content that request needs is sent to the AI provider that powers the feature. For the App's built-in features that is Anthropic, or a model reached through OpenRouter; if you use your own API key or choose your own AI provider, it is the provider you chose, under its terms. Some of these requests pass through Omniscio's AI gateway on Google Cloud in the United States, which forwards them without storing their content and records only usage amounts for billing. Where a provider offers a setting that stops it keeping or training on the data, Omniscio requests it. On new installs, Gmail and Calendar are left out of the daily digest until you turn them on; the digest never reads them while the Gmail or Calendar integration is switched off, and you can remove either from the digest in its settings at any time.
- Phone notifications. By default, new-email notifications on your phone contain only a reference to the message; the sender and subject are included only if you turn that on.
- Supermail's own sign-in. If you sign in to Supermail with its own Google sign-in instead of using Omniscio's Gmail connection, Supermail keeps your mail on its server (see §5). That sign-in asks Google for permission to read and change your mail (
gmail.modify), to send mail as you (gmail.send), and — when you open its send-as alias picker — to read your mail settings (gmail.settings.basic). - Things you choose to share. Content you publish, share or send goes where you send it.
Google user data is never sent to DeepSeek. While a Google account is connected, Omniscio does not run anything on DeepSeek on your computer: agent sessions cannot start or continue on a DeepSeek model (on any key or provider), and Ask Omniscio and the onboarding starter agents use a different provider. The built-in AI helpers never use DeepSeek at all. To use DeepSeek again you must first disconnect Google, which deletes the local Google data listed under "Your controls" below.
We do not sell Google user data, use it for advertising, or use it to decide creditworthiness. We do not use Google user data to develop, improve, or train generalized AI or machine-learning models. Omniscio staff do not read your Google data unless you ask us to (for example, in a support request), it is needed for security (such as investigating abuse), or the law requires it.
Your controls. You can disconnect Google at any time in Omniscio's Settings. Disconnecting revokes Omniscio's access at Google and deletes the local copy of your mailbox, your Google Meet list, the email samples Omniscio saved from Gmail, and the people and profile details it built from your email. Past daily digests, writing-style guides you created, and agent sessions you ran on your email — including bug reports that arrived by email — stay until you delete them. You can also remove Omniscio's access at myaccount.google.com/connections.
Limited Use. Omniscio's use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements.