Privacy Policy

GreatArrow.ai is a product operated by Manito AI, the parent company. This Privacy Policy describes how Manito AI (“we”, “our”, “us”) collects, uses, discloses, and protects information when you use our hosted Software-as-a-Service application at www.greatarrow.ai and related APIs (the “Service”). The Service was formerly hosted at greatarrowdigital.com, which remains a permanent alias for existing API and OAuth clients. It applies to personal information we process about you as the data controller; for content you store in your workspace (memories, files, agent outputs), we act as a processor on your behalf.

1. Information we collect

Account information

• Email address, hashed password (when you sign in with email + password), and OAuth identifier(s) when you sign in with a provider such as Google.
• Profile metadata you choose to add (display name, avatar URL).
• Workspace name(s) and the role(s) you hold in each.

Customer Content

• Memories, documents, chat sessions, agent runs, meeting transcripts, and any other content you upload, generate, or store. We process this on your behalf and do not access it beyond what is necessary to operate the Service.
• Coordination messages your connected agents send each other (for example, one agent asking another whether it is already working on a file). Encrypted at rest, private to you unless an agent addresses a message to a specific teammate or broadcasts it to the workspace, and governed by the “Agents asking each other” setting on your account page.
• Your career record. If you use the career search, the resume files, LinkedIn profile or data export, and the public web pages you name (your own website, articles about you) that you add to build your career record. We read each source once, keep only the roles, education, skills, links and press mentions it states — never the page text — and store the merged record in your workspace under the same access controls as your other content. You can remove any source or the record at any time under Your Work → Career. When you ask for a resume tailored to a job posting, what we write from that record — the tailored resume, a short cover note, and a PDF attached to the application card on your board — is stored the same way, in your workspace, under the same access controls. We never send any of it to an employer or a job board: you apply yourself, from the posting.
• Vector embeddings derived from your text. Stored alongside the source content under the same access controls.
• Conversation transcripts and summaries from connected AI clients. When you ask a connected assistant to save a conversation or a handoff checkpoint, the messages and summary it sends are stored on our servers in your workspace. You can read or delete them at any time under Your Stuff → Conversations, and deleting a conversation removes its stored messages.

Voice input and the “Hey Arrow” wake word

• Voice dictation and the optional “Hey Arrow” wake word are opt-in and OFF by default. Neither is active until you explicitly enable it in your account settings.
• When you enable the wake word, listening runs 100% on your device in your browser (an on-device neural model via WebAssembly). Audio is held only in a short rolling buffer in your browser’s memory and is never transmitted to, stored on, or processed by our servers — no audio leaves your device unless and until you explicitly trigger voice capture (by saying “Hey Arrow” or tapping the mic).
• What happens to a recording you do send. Once you trigger capture, that phrase is uploaded to your workspace’s private storage so it can be transcribed and so you can replay the message. We keep the recording for 30 days, after which it is deleted along with its metadata. Deleting the conversation deletes its recordings with it. We never use your recordings to train models.
• We may record an anonymized wake event (that a detection occurred, plus timing/accuracy diagnostics) as product telemetry. This telemetry contains no audio and no transcript of what was said.
• Because enabling the wake word turns on an always-on microphone on your device, we record your opt-in and any later opt-out in an append-only consent ledger, stamped with the version of this Policy in effect at the time. You can disable the wake word at any time from any surface (web or connected agent); disabling it stops all listening and records the withdrawal.

Calling or texting our phone number

• If you verify a phone number and call or text our number, Twilio transcribes what you say; we keep the transcript and our reply encrypted for up to 24 hours to answer the conversation, then delete them. We do not record call audio.

Text messages (SMS) we send

• Mobile information — including your phone number and your consent to be texted — is never shared with or sold to third parties or affiliates for their marketing or promotional purposes. We share a phone number only with the messaging carrier and with Twilio, our messaging provider, and only so a message you asked for can be delivered. Text-message originator opt-in data is not shared with any third party for any other purpose.
• Who we text, and why. We send SMS only to people who have added and verified their own phone number in their account settings. Three kinds of message are sent: a one-time verification code when you add a number; a reply from your own assistant, sent only in answer to a text you send us first; and operational alerts about your own account or workspace — a failed payment, a service incident, or a system alert, including production incident pages for staff administrators who enrol their own number. We do not send marketing or promotional text messages.
• How you opt in. You opt in by entering your phone number in Settings and confirming the verification code we send. No number is texted until that confirmation succeeds, and consent is never a condition of using Great Arrow.
• Message frequency varies and depends on your account activity — most people receive none in a typical week. Verification codes are sent only when you request one.
• Message and data rates may apply. Rates are set by your mobile carrier, not by us.
• Stopping messages. Reply STOP to any message to opt out; you will receive a single confirmation and no further texts. Reply HELP for help, or email support@greatarrow.ai. You can also remove your phone number in Settings at any time, which stops all SMS immediately. Carriers are not liable for delayed or undelivered messages.

Answering with a model on your own computer

• Running answers on your own computer (via Ollama or LM Studio) is opt-in and OFF by default, and applies only to accounts where we have enabled the feature.
• It does not change what we collect. Your question is still received, checked, and stored by us exactly as it is when our own AI answers, and the context used to answer it still comes from retrieval we perform on our side. What changes is only which machine writes the answer. Do not read this as the wake-word guarantee above: that one is about audio, and it does not apply to your question here.
• We store the name of the model you selected (and the list of models your browser reported), so a later request can be checked against it. We never download, run, or receive the model itself.
• Answers written on your own computer cannot reach your connected accounts. You can switch back to our AI at any time from any surface (web or connected agent).

Integration credentials

• OAuth refresh / access tokens for third-party providers you choose to connect (Google Workspace, Microsoft 365, Slack, Notion, Atlassian, Asana, Discord, Zoom, Dropbox, GitHub, GitLab). Encrypted at rest with AES-256-GCM under a key we control or a per-workspace key (BYOK) if you provide one. Decrypted in-process only for the duration of an integration sync.

Databases you connect

• When you connect a database (today, a Supabase project), we read it read-only, and only the tables you select. Every query is checked to be a read before it is sent and runs inside a read-only transaction, so nothing in your database can be changed from here.
• For charts, we read the columns you mark as report columns, plus row counts and dates. Columns whose names suggest personal or secret values — email, phone, address, tokens, keys — are refused as chart columns even if you mark them.
• A chart is shown on a page you share with a client only when you turn on Share with client for that table. Turning it off holds the chart back the next time the page is opened.
• Rows are remembered — copied into your workspace so they can be searched — only for tables where you turn that on separately. It is off by default.

Operational data

• Audit log: every read and write against your workspace, including actor, IP address, user-agent, and resource. Used for security incident response and for the in-app audit dashboard.
• Signup record: when an account is created we record the sign-in provider, the domain part of the email address (never the address itself), the IP address, the country your network resolves to, and the browser user-agent. Used only to tell genuine sign-ups from automated abuse. It records; it never blocks a sign-up.
• Usage metrics: token counts, model identifiers, latency, and error rates per request — used for billing, capacity planning, and SLO monitoring.
• Diagnostic data: error reports, stack traces, and performance spans (Sentry, Vercel logs). Personal information appearing in stack traces is redacted at our logger boundary where possible.
• Optional CLI, mobile, and ArrowDeck desktop diagnostics: when you explicitly enable diagnostic telemetry, we receive redacted error events, client version, platform, and limited technical context. Credentials, cookies, contact fields, and message contents are filtered before transmission. Mobile diagnostics are buffered locally only while consent is active; CLI diagnostics can be disabled withGAD_TELEMETRY=0 or the CLI configuration. ArrowDeck diagnostics additionally include an install identifier — an opaque value the app generates on first run so we can tell how many installations are affected by a fault rather than only how many error reports arrived. It is not derived from your hardware, machine name, or account, it is not linked to your Great Arrow user, and reinstalling the app produces a new one. We do not store your IP address with these events. All diagnostic events, and the install identifier with them, are deleted after 90 days, and are used solely for reliability and incident response.

2. How we use information

• To provide and maintain the Service for you.
• To authenticate you, enforce workspace boundaries, and prevent abuse.
• To bill you, manage subscriptions, and resolve billing disputes.
• To respond to support requests, communicate Service updates, and send transactional email (invitations, password resets, health digests).
• To detect, investigate, and prevent fraud, security incidents, and abuse.
• To comply with legal obligations and respond to lawful requests by public authorities.

We do not sell, rent, or trade personal information. We do not use Customer Content to train large language models for any third party, and we do not use it for our own model training without your separately-given consent.

2a. Legal bases (EEA / UK)

If you are in the European Economic Area or the United Kingdom, we process your personal information on the following lawful bases:

• Performance of a contract — providing, maintaining, and securing the Service, authenticating you, and billing you.
• Legitimate interests — preventing fraud, abuse, and security incidents, and keeping the Service reliable, provided these interests are not overridden by your rights.
• Consent — optional analytics cookies (see the Cookies & tracking section) and any use of your Customer Content to improve our own models. You may withdraw consent at any time.
• Legal obligation — retaining billing/tax records and responding to lawful requests by public authorities.

2b. AI processing

• When you ask an AI feature to act on your content, the specific content you reference is transmitted to our configured inference providers (Anthropic, OpenAI, or Google via the Vercel AI Gateway, or an upstream provider selected by OpenRouter with data_collection: deny) solely to produce your requested output. This applies to all Customer Content you direct an AI feature to process, not only Google data.
• Bring-your-own-key (BYOK): if you supply your own model provider key, calls made with it are routed to and billed by your provider under their terms; we do not incur the model cost and the provider’s data-use terms govern that call.
• Model training: we use data to improve our own routing and models only where you have given machine-learning consent and the data is content you authored — never content synced from a connected integration, and never data received from Google APIs.

2b-ii. Text messages and phone calls (opt-in)

You can add a mobile number under Account → Phone so that calling or texting the Great Arrow number reaches your own assistant. This is opt-in and OFF by default, and is included only on paid plans.

• What we send: a one-time verification code when you add your number, and replies from your assistant to messages you send first. We never send marketing or promotional texts.
• Consent: you tick a checkbox agreeing to receive these messages before a code is sent. Message frequency varies with your own use. Message and data rates may apply. Reply STOP to opt out at any time and HELP for help; removing the number in Account → Phone has the same effect.
• Calls: calls to the Great Arrow number are transcribed by Twilio so your assistant can answer. The transcript is held encrypted for up to 24 hours to carry the conversation, then deleted. We do not record call audio.
• No sharing: your mobile number and SMS consent are never sold, rented, or shared with third parties or affiliates for marketing or promotional purposes. Twilio processes messages and calls on our behalf as a service provider.
• Limits and cost: each plan includes a monthly allowance of call minutes and text replies. When it is used up the channel pauses until the next month; nothing is billed beyond your plan.

2b-i. Product-update email (opt-in)

We send two different kinds of email, and they are governed differently.

• Service email is not optional and is sent on the basis of performing our contract with you: sign-in links, password resets, receipts and billing notices, workspace invitations, security alerts, and the daily digest of your own workspace you asked us to send. Your daily digest has its own settings and its own unsubscribe link at /account.
• Product-update email is opt-in and OFF by default. This is commercial email about new features and changes to the product. We send it only where you have affirmatively opted in — either by ticking the separate, unticked box at sign-up or by turning on “Product updates” at /account. Our lawful basis is consent.

Every opt-in and every opt-out is recorded in our consent ledger with the time, the policy version you were shown, and which surface you used, so we can demonstrate that consent was active on the day we sent. You can withdraw at any time — with the toggle at /account, or with the one-click unsubscribe link that appears in the footer of every product-update email. Both routes take effect immediately and identically. Withdrawing consent for product updates does not affect the service email described above.

2c. Model fine-tuning program (opt-in)

We are introducing an opt-in program to fine-tune our own models on consented Customer Content. It works as follows:

• Opt-in only: your content is eligible only while your machine-learning consent (the “Improve our models” setting at /account) is switched on — for team workspaces, both the workspace owner’s and the content author’s consent are required. Withdrawing consent removes your content from future training runs and schedules previously collected training data for deletion.
• What is eligible: only content you authored (memories and chats you wrote in the Service). Content synced from connected integrations — including anything received from Google APIs — is never eligible, regardless of consent.
• Redaction: eligible content is stripped of workspace and user identifiers and passed through automated redaction of emails, phone numbers, government identifiers, payment card numbers, IP addresses, and street addresses before it is used. This is pseudonymization, not anonymization.
• Fine-tuning compute: training runs execute on Thinking Machines Lab (the Tinker training service), acting as our sub-processor (see Section 3). Only the redacted dataset is transmitted; prepared datasets are deleted after each run, and the resulting model adapters are used solely to improve the Service and are never shared or sold.
• Timing: this disclosure is published on the revision date above. Consistent with our 30-day change-notice commitment (Section 10), no customer content will be used for fine-tuning before 2026-08-19. You can withdraw consent at any time before or after that date.

2d. Product feedback you submit

When you use the in-app Report tool to tell us about a bug, idea, question, or task, that submission is used to improve Great Arrow itself. Because it is feedback about our product rather than your own workspace data, the report is delivered to our own product-development workspace — it leaves the workspace you filed it from, and our team can read it.

A report includes what you typed plus context we attach to make it actionable: the page you were on, the on-screen elements you pointed at (including their visible text), your browser, screen size, theme, language, timezone, recent in-page errors, your name, and the name and id of the workspace you filed from. If you choose to attach a screenshot, that image is a picture of your screen and may therefore contain your own Customer Content. Attaching a screenshot is always optional, and you can remove it before submitting.

Submitted text and screenshots are scanned for malicious content before they are stored. We use reports only to diagnose and improve the Service; we do not use them for advertising, and we do not sell them. The retention and deletion rights in Sections 4 and 6 apply.

3. Sub-processors

We rely on the following sub-processors to operate the Service. Each operates under written terms that include confidentiality and security obligations no less protective than those in this Policy:

We will provide at least thirty (30) days’ notice before adding a new sub-processor or materially changing the role of an existing one. To object, contact privacy@greatarrow.ai.

3a. Google API Services — Limited Use

GreatArrow.ai’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. We do not sell Google user data, do not use it for advertising, and use it only for the user-facing features described below.

We do not use information received from Google APIs to develop, improve, or train generalized AI and/or ML models.

When you connect Google integrations we may access:

• Gmail: send-only access to send messages on your behalf (gmail.send). We do not read your mailbox: no Gmail read scope is requested, so inbox contents, message metadata and attachments are never accessed or ingested.
• Google Drive: access limited to the files you explicitly pick and the files this app creates (drive.file). We do not have read access to your Drive as a whole.
• Google Calendar: read and write access to view, create, update and cancel events and generate summaries.
• Google Contacts: read-only access to names, emails and birthdays for meeting scheduling and reminders.
• Google Docs / Google Sheets / Google Slides: read-only content extraction; Google Sheets additionally supports creating and writing spreadsheets you request.
• Google Analytics: read-only GA4 reporting for analytics summaries.

When you ask an AI assistant to act on this data (for example, “summarize my inbox”), the specific content you reference is transmitted to our configured AI inference providers solely to produce your requested output. By default these are Anthropic, OpenAI, and Google. Cost-optimized routing may instead serve a request through OpenRouter, Inc., which forwards it to an upstream provider it selects; we configure OpenRouter to route only to providers that do not collect or train on your data (data_collection: deny). None of these providers train language models on your data per their usage terms and our agreements.

You can revoke access at any time at /account/connections, which deletes stored credentials within 24 hours and revokes the token upstream, or delete your entire account at /account.

4. Retention

• Account & profile data: retained for the duration of your account; deleted within thirty (30) days of account closure.
• Inactive accounts: if an account goes twenty-four (24) consecutive months without activity, we may close it and delete its data on the schedule above. Activity means any sign-in or any use of an API token or connected agent — an account used only through the MCP interface is an active account. We email the address on the account at least thirty (30) days before any such closure, and keeping the account takes one click. We do not close an account with an active paid plan, and we do not close an account that is the sole owner of a workspace other people belong to.
• Customer Content (memories, documents, chats): retained while you maintain a workspace. Deleting a workspace soft-deletes its content; hard deletion completes thirty (30) days later. Encrypted database backups may persist for up to ninety (90) days.
• Integration credentials: deleted within twenty-four (24) hours of disconnect; we attempt upstream revocation where the provider supports it.
• Wake-word audio: not retained. The “Hey Arrow” rolling buffer lives only in your browser’s memory and is discarded continuously; no wake-word audio is ever sent to or stored by us. Anonymized wake-event telemetry (no audio) follows the operational-logs window below.
• Signup records: retained for one hundred and eighty (180) days, then deleted. The window is set to outlast an abuse investigation rather than to accumulate history.
• Audit log: retained for ninety (90) days.
• Operational logs & metrics: retained for thirty (30) days, except aggregated, non-identifying usage counters which we retain longer for capacity planning.

5. Security

• All traffic is encrypted in transit via TLS 1.2 or higher.
• Personal access tokens are stored hashed (SHA-256). The plaintext is shown to you exactly once at mint time.
• OAuth refresh tokens are encrypted at rest with AES-256-GCM. Per-workspace BYOK is available.
• Workspace data isolation is enforced in our application layer: every query that touches workspace data carries the workspace you are signed in to, resolved on our servers from your session and never from anything your browser sends. A build-blocking test fails our release if any query reaches workspace data without that scope. Postgres row-level security is enabled on every workspace-scoped table as an additional safeguard against direct database access.
• Realtime broadcast channels use HMAC-signed names so a stranger who learns a workspace ID cannot subscribe to its events.
• We follow least-privilege for operator access; admin actions are audit-logged.
• Breach notification: if a personal-data breach that affects you occurs, we will notify affected users and the applicable supervisory authorities without undue delay as required by law.

6. Your rights

Depending on your jurisdiction you may have rights to access, correct, export, delete, restrict, or object to processing of your personal information, and to lodge a complaint with a supervisory authority. To exercise these rights:

• Access & export: use the in-app export at your workspace settings, or email privacy@greatarrow.ai.
• Correct: update profile fields from /account.
• Delete: close your account from /account; revoke a single integration from /account/connections.
• Object / restrict: email privacy@greatarrow.ai and we will respond within the timelines required by applicable law.

6a. California privacy rights (CCPA / CPRA)

If you are a California resident, you have the right to know the categories of personal information we collect (see Section 1) and the purposes for which we use them (Section 2), and the right to access, delete, and correct that information (Section 6). We do not sell your personal information. We also do not “share” it for cross-context behavioral advertising; the only third parties that receive analytics data are Microsoft Clarity, Google Analytics, and Ellm (marketing-homepage referral attribution only), all of which load solely when you opt in via the cookie banner and which you can turn off at any time in Privacy Settings, and our hosting provider Vercel, whose cookieless Web Analytics counts page views (path only, no query string, no identifier stored on your device) through a separate per-page request to Vercel itself, as described in section 3.

To exercise these rights, or to appeal a decision, email privacy@greatarrow.ai. We will not discriminate against you for exercising them. You may use an authorized agent to submit a request on your behalf.

7. International transfers

Data is processed in the United States during the current rollout. If you are located in the European Economic Area, the United Kingdom, or Switzerland, transfers of personal information to the United States are made under appropriate safeguards (such as the Standard Contractual Clauses approved by the European Commission). Multi-region replicas are on the roadmap; until then, please do not store data you are required by law to keep within another jurisdiction.

8. Children

The Service is not intended for, and we do not knowingly collect information from, children under thirteen (13) — sixteen (16) in the EEA / UK. If you believe we have collected such information, contact us at privacy@greatarrow.ai and we will delete it.

9. Cookies & tracking

We use a small number of cookies and similar technologies:

• Essential cookies (always on): your sign-in session, the workspace you have selected, and your cookie choice itself (gad_cookie_consent). These are required for the Service to function and are not used for tracking.
• Optional analytics (off by default): Microsoft Clarity and Google Analytics, used to understand product usage, and Ellm on the marketing homepage, used to attribute which AI assistant or search engine referred a visit. These load only after you choose “Accept all” on the cookie banner. Ellm sets no cookie or other identifier on your device; it is gated here because it reports to a third party.
• Vercel Web Analytics (always on, cookieless): our hosting provider counts page views so we can see which public pages lead to a sign-up. A script on the page sends Vercel one request per page view with the page path (query string already removed), the referring site, a coarse IP-derived location, your operating system, browser and device type; it sets no cookie or other identifier on your device, and it sends nothing from the sign-in callback, workspace-invite, share-link or agent-pairing pages. It is not gated by the banner because it reports to no additional party: Vercel already receives the same request details as our hosting provider in order to serve the page. The full data list is in section 3.

You can change your choice at any time from the “Privacy Settings” / cookie-preferences control in the app; declining leaves only the essential cookies in place. We honor your selection across the Service.

10. Changes to this Policy

We may update this Privacy Policy from time to time. Material changes will be communicated by email or via the Service at least thirty (30) days before they take effect. The current revision date appears at the top of this page.

11. Contact

Privacy: privacy@greatarrow.ai
Security: security@greatarrow.ai
See also our Terms of Service.