Harmonyk — Privacy Policy
Last updated: 2026-11-06 · Version 1.0
Harmonyk is operated by Harmonyk LLC, a Wyoming limited liability company ("Harmonyk", "we", "us"). This policy explains what personal data we process, why, who else processes it, and what rights you have. It applies to the Harmonyk application and the harmonyk.ai website.
If you use Harmonyk through your employer or another organisation, that organisation decides why your workspace content is processed and we process it on their behalf (see §1). Where we decide how and why to process data ourselves — your account, billing, security, and running the service — we are responsible for it under data-protection law.
1 · Who is responsible for what
We play two roles, and we keep them apart:
- Processor. For workspace content — documents, extracted text, versions, Rooms, Maestro conversations held inside your workspace, and the receipts tied to that content — we act as a processor on behalf of the organisation that owns the workspace. That organisation is the controller of that content. We process it only to provide the service and only on their instructions.
- Controller. For account data (email, password, signup records), billing and metering records, security and abuse logs, product usage patterns, and support correspondence, we act as controller — this data exists because we operate the service as a business.
If you believe workspace content of yours is being processed in a workspace you don't control, contact us (§13) and we will route the request to the workspace owner.
2 · What we collect, and why
| Data | What it is | Why we hold it | Lawful basis (GDPR) |
|---|---|---|---|
| Account data | Email, password (handled by our authentication provider — we never store it in plaintext), signup consent record (timestamp + version of the terms accepted) | Create and secure your account; prove what you agreed to | Contract; legal obligation |
| Workspace content | Documents, extracted text, versions, Rooms | The product itself: store, index, and version your documents | Contract (processed on behalf of the workspace owner) |
| Maestro conversations | Your messages to Maestro, Maestro's answers, action cards and receipts | Answer you; keep the record of what was asked and done | Contract; consent for AI processing (§3) |
| Receipts & manifests | Import review records and evidence lines showing what was read, imported, or changed | Our proof of care: when you ask "what happened to my data", we answer with records, not assurances | Contract; legitimate interests (accountability) |
| Usage & metering | Metering events, quota counters, estimated cost | Meters you can see; billing; abuse detection | Contract; legitimate interests |
| Logs & security data | Standard server and authentication logs, rate-limit state | Keep the service safe and available; prevent abuse | Legitimate interests |
| Product usage patterns | Which features you use and how often (observed product behaviour) | Decide what to build and automate; improve the product | Legitimate interests |
| Connector data (if you enable a connector) | OAuth tokens, account names, the data you selected for import, per-import receipts (§3) | Run exactly the connectors you granted, and show you what they did | Consent — specific, revocable (§3) |
| Maestro memory | Content retained by Maestro from past conversations to give continuity | Better continuity of answers | Contract; you can turn it off (§6) |
We do not collect special-category data on purpose. Because the product holds whatever documents you put in it, sensitive content may end up in workspace content and Maestro conversations. It is protected the same way as everything else and reaches AI processors only as described in §3.
3 · How your content flows through AI processors
One design law governs everything in this section: only bounded extracted text leaves the application to a model provider, only on routes configured for no data collection and zero data retention, and if no such route exists the request fails rather than falling back. File extraction runs on our servers; the original file bytes of your documents are never sent to a model provider.
In full:
- Embeddings and semantic search. Document text and every search question are sent to Cohere to create embeddings and retrieve results. Cohere deletes logged prompts and outputs after 30 days, and we keep its training-on-prompts setting switched off in our account: your content is not used to train Cohere's models (§4).
- Generation. Your prompts and the extracted context needed to answer go to OpenRouter, which routes them to model providers, on zero-data-retention / no-data-collection routes only. OpenRouter does not retain prompt or response content on these routes; it records request metadata only (timestamps, model, token counts).
- Maestro. To answer your questions, Maestro reads your workspace — documents, Rooms, tasks, settings. That is the feature: answers grounded in your documents. What Maestro reads and does is receipted like everything else.
- Web research. When a question needs current information, Maestro may fetch public web pages to answer it. Only the question and the URLs needed to answer are involved.
- Connectors (Gmail, Google Drive). If you enable a connector, we access your account read-only, through the OAuth scopes you grant, and only import what that connector is set to import: for Gmail, message metadata and — unless you turn it off — message text; for Drive, file metadata. Each import is reviewed against a manifest before it enters your workspace and is receipted afterwards with its evidence line. We never send, label, move, or delete anything in your mailbox or Drive.
We do not use your content to train AI models. For generation this is enforced by the zero-data-retention route policy above; for embeddings by our account settings at Cohere.
When you upload content for AI processing, the product asks for your consent first and records it. You can withdraw consent by not using AI features on that content and by contacting us (§13).
4 · Who processes your data — the sub-processor register
These are the third parties that process personal data to run Harmonyk. A new sub-processor is named here before it touches any data (§5, §12); this register is versioned together with the policy.
| Sub-processor | Purpose | What they see | Region | Retention |
|---|---|---|---|---|
| Supabase | Authentication, database, file storage — the workspace itself | Everything the product stores (content, account, metering) | United States (us-east-1) | Kept for as long as your workspace exists; deletion per §6 |
| OpenRouter | LLM generation (Maestro answers, drafts) | Prompts + bounded extracted text — zero-data-retention / no-data-collection routes only | United States (request routing) | No retention of prompt/response content; request metadata only |
| Cohere | Embeddings + retrieval for semantic search | Document text + search questions | United States | Logged prompts/outputs deleted after 30 days; not used for training |
| Vercel | Hosting / compute | Request data, application logs | United States (global edge network) | Platform logs retained briefly, per Vercel's terms |
| Resend | Transactional email (signup confirmation, product notices) | Recipient email, email template content | Emails sent from Tokyo (ap-northeast-1); Resend account data and logs stored in the United States | Per Resend's terms |
| Gmail/Drive connectors (only if you enable them) | Only what the connector is set to import (§3) | Google's regions | Per Google's terms; removed from your workspace when you delete it (§6) |
Payments. We do not currently use a payment provider: plans and quotas are managed manually by us. We do not collect payment card details. If that changes, the provider will be named in this register before it processes anything.
5 · What we deliberately do not do
- We do not sell your data. Not to data brokers, not to advertisers, not to anyone.
- No advertising, no ad targeting. The product has no advertising business model.
- No training on your content (§3).
- No surprise processing. Every new sub-processor is named in §4 before it touches data, and additions are announced (§12).
- The receipts, manifests, and evidence lines described in §2 exist so that these promises can be checked, not just believed.
6 · Retention & erasure
What we keep, and for how long. Workspace content: as long as the workspace exists. Account data: for the life of your account. Security and abuse logs: as long as needed to protect the service and investigate incidents, then deleted or anonymised. Metering and billing records: as long as needed for billing, accounting, and legal obligations. Maestro memory: on by default; retained from your conversations until you turn memory off or ask us to clear it (below).
Memory. Maestro keeps a rolling memory of your recent messages to give answers continuity. You can turn memory off in Maestro's settings; that stops future retention. Memory already stored is not purged automatically — ask us (§13) and we will clear it.
Append-only history. The product's history is append-only by design: Rooms can be archived and renamed but not deleted, because the record of what was agreed is the point of the product. This means erasure requests that reach workspace content are handled case by case, together with the workspace owner: we will tell you plainly what can be removed, what must be kept, and why. Where we can destroy content without breaking that record, we will.
Account deletion. Ask us to delete your account (§13) and we will handle it manually and confirm when it is done — within one month at the latest, and sooner where we can. Records we must keep for legal or accounting purposes (for example, billing records) are kept only as long as those requirements demand, then deleted too.
Free tier. The free plan's document allowance is 250 documents for the lifetime of the workspace; deleting documents does not restore quota.
Receipts quoting your content. A receipt may quote a small excerpt of content — for example, from your own connected mailbox — as its evidence line. Erasure covers receipt excerpts too: we redact or destroy them along with the underlying content.
7 · International transfers
Harmonyk LLC is established in the United States, and our sub-processors process data in the United States or route it globally (§4). If you use Harmonyk from the EEA, the UK, or Switzerland, your data will be transferred to and processed in the United States and other countries that may not have the same data-protection rules as yours.
Where such transfers are restricted, we rely on the safeguards our sub-processors commit to in their data-processing agreements — in practice the EU Standard Contractual Clauses (and the UK International Data Transfer Addendum where applicable) plus each provider's documented supplementary measures. Note that Resend's sending region only affects where email is dispatched from (currently Tokyo); Resend stores account data and logs in the United States regardless of sending region.
8 · Your rights
Depending on where you live, you may have the right to: access your personal data; rectify it; erase it; restrict processing; object to processing based on legitimate interests; portability; and withdraw consent at any time, without affecting processing already done.
To exercise any of these, email reuben@harmonyk.ai. We answer within one month and may ask you to verify your identity. Subject-access requests can be answered from the receipt and manifest records — the product is designed to answer "what happened to my data" with rows.
If you are not satisfied, you can complain to your local data-protection supervisory authority — for example Datatilsynet in Denmark, the CNPD in Luxembourg, or the ICO in the UK, depending on where you live or work.
9 · Security
Measures include: authentication handled by our authentication provider (Supabase), so we never store your password; all traffic encrypted in transit (HTTPS); connectors accessed only through OAuth — we never see your Google password; per-upload consent gates before content is sent to AI processors; and access controls limiting who at Harmonyk can reach production data.
No system is perfectly secure. If we learn of a breach affecting your personal data, we will notify you as required by law and by our contract with you.
10 · Cookies & similar technologies
The harmonyk.ai website sets no cookies and runs no analytics or advertising trackers — there is nothing to consent to. If that ever changes, this section and the register in §4 change with it, with notice (§12).
Inside the application we use token-based authentication (via Supabase) and necessary local storage to keep you signed in and run the app. These are strictly necessary for the service — no consent is required for them, and they are not used for tracking or advertising.
11 · Children
Harmonyk is a business tool and is not directed at children. You must be at least 18 to use it. We do not knowingly collect data from children.
12 · Changes to this policy
We version and date this policy. Material changes — including any addition to the sub-processor register in §4, and any change to §10 — are announced to workspace owners by email or in-product notice before they take effect, and we keep previous versions available so you can compare.
13 · Contact
Harmonyk LLC (Wyoming, USA). Privacy questions, requests, and erasure/deletion requests: reuben@harmonyk.ai.
This policy is written in plain language on purpose. If anything here is unclear, ask us — we would rather explain than have you guess.