Privacy policy

This policy explains how personal data is processed when you use Entumo, in accordance with the revised Swiss Federal Act on Data Protection (revDSG) and its ordinance (DSV). Where the EU GDPR also applies, we observe it.

1. Controller

The controller responsible for the data processing described here is MomentumQ GmbH, Leutschenbachstrasse 95, 8050 Zürich, Switzerland. Data-protection enquiries: info@momentumq.com, or by post to the address above.

When your agents process personal data of your customers, employees or contacts, your company is the controller of that data and we act as its processor (terms, section 12). People whose data your workspace processes should address their requests — including a request that a person review an automated decision — to your company; we help you answer them, and workspace owners have a search that finds one person’s data across the workspace.

2. What data we process

• Account data — name, email address and a hashed password; workspace memberships; acknowledgments you give at sign-up or when joining a workspace (such as the AI-use acknowledgment), with their timestamp and the terms version then in force; each acceptance of the terms is also kept as a record of its own (see section 7). If you use Sign in with ChatGPT (available only for an existing account), OpenAI tells us your e-mail address, whether it is verified, your name and your OpenAI account identifier; we use the address to find your account and store only the identifier with it, so that we recognise you at later sign-ins, until your account is deleted (section 4).
• Security data — if you enable two-factor sign-in, an encrypted authenticator secret and hashed recovery codes; a short-lived record that you recently re-entered your password before a sensitive action. If you register a passkey: its public key and credential identifier, the name you give it, its device type and whether it is backed up, the authenticator model identifier, a signature counter, and when it was created and last used. The private key never leaves your device.
• Session data — a random session identifier in a cookie; our database stores only a digest of it. A session ends after 12 hours without activity and at the latest after 48 hours.
• Usage / log data — IP address, timestamp and browser user-agent, and for sign-in, sign-up, password-reset and similar account events also your e-mail address or user id; these exist only in our hosting provider’s log stream (for security and troubleshooting) and are deleted there automatically within at most 30 days. To limit abuse, our database also counts requests per IP address and sign-in attempts per email address for a short window (one minute to one day). These counters are stored under a keyed digest, never the address itself, and are deleted at the latest a day after their window ends.
• Workspace settings — the company profile, the email signature (which can contain a person’s name, role and phone number), invitations, and the notification targets you set (email address, WhatsApp number, Threema ID); notifications are delivered through the provider of the channel you choose; if an owner enables the China data region (section 5), we also keep a record of that confirmation — who, when and the wording shown.
• Billing data — plan, Stripe customer and payment references, the credit ledger and usage per AI model. Card and bank details are held by Stripe, not by us. If an owner switches on automatic top-up, we also keep the record of that agreement — who confirmed it, when, the wording and the amounts and daily limits — with the wallet.
• Content you provide — task briefs, messages, files your AI team produces, and the connector credentials you add (stored encrypted; see “Security”).
• Voice recordings — when you dictate, the audio is sent for transcription to a provider listed in section 4 and is not kept afterwards; the transcript lands where you dictated it.
• Face-to-face conversation — if you open the optional face-to-face conversation, your microphone audio and, if you switch it on, your camera video are streamed live to the video provider named in section 4 for the length of the conversation, together with a fixed instruction and the language of your screen. Your voice and face can identify you and may be sensitive data; the conversation therefore opens only after you have read the notice and pressed Start (section 3). We send nothing from your workspace and store no audio, video or transcript; we keep only the fact that a conversation took place and its charge (section 7).
• Customer data you process through Entumo — when an AI agent acts in a connected tool (e.g. an inbox, accounting or payment system), it processes the data in that tool on your behalf; for that data you are the controller and we act as your processor.
• Product statistics — events such as “task completed” or a finished onboarding step, linked to your user and workspace identifiers but without any content, so we can see how the product is used and improve it; the sign-up and access-request events also record the domain of the e-mail address used (for example example.ch), never the address itself. They are kept for 180 days.
• Partner referral — if you arrived through a link from one of our partners (for example a fiduciary), the partner code in that link is stored with the workspace you create, and that partner sees your workspace’s name and creation date in their client list.
• Abuse reports — if you use the public report form or write to our abuse address: the category, the link and description you give, the e-mail address if you leave one, and the time of the report; we use them to handle the report, to contact you with questions and to establish, exercise or defend legal claims. We keep a report only as long as that requires and, on request, delete your contact details from it unless they are needed for a pending case.
• Recipients of Entumo-hosted links — e-mails your agents send can show your signature logo through a link to our server; when a recipient’s mail program loads the image, it sends its IP address, the identifier of its browser or mail program and the time to us and to our hosting provider, and they appear in the log stream described above. The link is the same for every recipient of your workspace, so we cannot tell recipients apart, and we do not use the requests to track anyone.

Marketing measurement. On our public pages your browser stores, in session storage on your device, the campaign parameters in the link you arrived through (utm_source, utm_medium, utm_campaign), a partner code if the link carried one, the referring website’s host name, the page you landed on and which button you pressed. No cookie is set, and it is gone when you close the tab. If you then send us an access request or sign up, that context travels with the form so we can see which channel works (at sign-up it is kept with the product statistics above). In addition, the first view of a public page in a tab reports the page, the language and the referring host to our server, which keeps only an aggregate count per page and day (for 365 days, with no identifier); to stop strangers inflating those counts, the server counts views per IP address for one hour, under a keyed digest rather than the address. You can refuse all of this by blocking site data for this domain in your browser settings; everything else on the site keeps working. We do not use it to recognise you on other websites.

Access requests. If you ask for access through the form on our website, we store your email address, company, the task you describe, the plan and language you chose and the campaign context above, and send a copy to our own mailbox, in order to answer you and plan your onboarding. We keep the request for 12 months.

3. Purposes and legal bases

We process the data to provide and secure the service and manage your account (performance of our contract with you), to operate, improve and safeguard it against abuse, to measure which channels and features work and to settle partner arrangements (our legitimate interests, weighed against your rights — you may object at any time; see “Your rights”), and to meet legal obligations such as statutory accounting retention. Apart from the optional face-to-face conversation described below, we do not rely on consent for this processing; where we ever ask for your consent, you may withdraw it at any time with effect for the future. Providing the data is voluntary, but without it the service cannot be used.

Face-to-face conversation. Your voice and face can identify you and may count as sensitive personal data, so for this optional feature we do rely on your consent: before anything is sent we show what is sent, to whom and what a conversation costs, and the conversation opens only when you press Start. You can end it at any time, which withdraws your consent with effect for the future; what the provider already holds is dealt with in sections 4 and 7. If someone else will be heard or seen, do not open it without their agreement. The feature is offered only where your workspace’s data-residency policy allows the USA and its provider scope is not restricted to the providers its employees are assigned to (Security page).

4. Disclosure to processors and third parties

To run the service we rely on the following processors, who act only on our instructions. Each entry names the state in which the processor handles the data and the basis of that disclosure (see section 5).

• AI model providers (to generate the work): your employees run on the models you assign to them — Anthropic (Claude) by default, or the European versions below in an EU-hosted workspace. The manager — the assistant that answers in your workspace chat and sets up work — runs by default on OpenAI (GPT-6 Sol) where that provider is enabled on our deployment and your residency and provider settings allow it (in an EU-hosted workspace on OpenAI’s European region or Azure’s EU data zone), and otherwise on Anthropic (Claude) or another provider from the list; an owner can choose another manager model in Settings. The manager receives the content of your conversation with it and what it reads in your connected tools. The platform also uses providers from this list for supporting tasks, even if you have not assigned work to their models: summarising long tool results and conversation histories, naming chats, planning a brief, checking a task and an outgoing message, reading scanned documents and images, transcribing dictation and reading replies aloud. Those providers receive the content concerned, which can include emails, invoices and other documents your agents read. Your data-residency policy applies to every one of these calls — a provider in a region you have not allowed receives nothing — and you can switch off individual providers in the spend controls. With the setting “only the providers my employees are assigned to” on the Security page, every supporting task stays with the providers of your employees’ models; a supporting feature that none of them can serve is then unavailable. The setting is off by default, only an owner can change it, and it is suggested for professions bound by secrecy; it also covers the manager and turns smart model routing off. Each route counts on its own (Claude through Anthropic does not bring in AWS Bedrock), an assist model an owner picked in Settings counts as assigned, and web search (Tavily) and connected apps such as DeepL are not AI model providers and follow each employee’s tool access. The China-region models (Kimi, Qwen, Z.AI) run only if you first enable the China region on the Security page. Providers marked “coming soon” in the product cannot be enabled yet. The complete list — identical to the sub-processor list on the in-app Security page:
  • Claude — Anthropic Ireland, Limited and Anthropic PBC, USA or worldwide (the provider may route requests globally) (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • Kimi — Moonshot AI, China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  • OpenAI — OpenAI Ireland Limited and OpenAI, USA (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • Gemini — Google, USA or worldwide (the provider may route requests globally) (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • Grok — xAI Corp., USA (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • Mistral — Mistral AI, EU (adequate level of protection (Annex 1 DSV))
  • AWS Bedrock — Amazon Web Services (Bedrock, EU regions), EU and Switzerland (Frankfurt, Zurich, Stockholm, Milan, Spain, Ireland, Paris) (adequate state; SCC for access from the provider's US group)
  • Azure Foundry — Microsoft (Azure AI Foundry, EU data zone), EU, may include Switzerland and Norway (adequate state; SCC for access from the provider's US group)
  • OpenAI EU — OpenAI (European data residency, eu.api.openai.com), EEA and Switzerland (adequate state; SCC for access from the provider's US group)
  • Vertex AI EU — Google Cloud (Vertex AI, EU multi-region), EU (adequate state; SCC for access from the provider's US group)
  • Azure Foundry (global) — Microsoft (Azure AI Foundry, global routing), worldwide (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • NVIDIA NIM — NVIDIA Corporation (NVIDIA NIM), USA (FDPIC-recognised SCC; Swiss-US DPF where certified)
  • Qwen — Alibaba Cloud (intl: Singapore), Singapore (storage); processing on international nodes worldwide outside mainland China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  • Z.AI — Z.AI / Zhipu (Beijing), China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  • DeepSeek — DeepSeek (Hangzhou), China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  • MiniMax — MiniMax (Shanghai; intl endpoint), China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  • Xiaomi MiMo — Xiaomi (MiMo API), China (no adequacy, no SCC — only with your Art. 17 DSG basis)
  With the daily knowledge fold enabled (the default), a clipped summary of the day’s completed work is additionally distilled into your workspace’s own knowledge base by an AI model chosen under the same provider and residency rules — you can turn this off in Settings. With the optional daily log switched on in Settings (off by default), a provider of the same kind receives clipped briefs and outcomes of the day’s runs to write a readable summary for the owner. When you choose «Fill from the web» for your legal company details, OpenAI (USA) receives the company name and the website, UID and address you provide to research public sources. This optional lookup follows your workspace’s model and residency rules; researched values are saved only when you save the settings.
• Connector providers (only the ones you connect, and only for the actions you authorise): e.g. Bexio, Stripe, Slack, Microsoft 365, Google Workspace, WhatsApp, Threema and others listed on the Integrations page. For Google Workspace connectors (Gmail, Google Calendar, Google Drive), Entumo’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including its Limited Use requirements — in particular, we do not retain, use or transfer data obtained through Google Workspace APIs to develop, improve or train generalized (non-personalized) AI or machine-learning models, and we do not transfer it for advertising. What an employee reads from a Google service or does with it stays in that task’s run history in your workspace until you delete the run or the workspace, and what it reads goes to the AI model provider that runs it (see above); the app shows you this before it sends you to Google. For the X connectors (posts and advertising), agents act on your X account and X Ads account only after you sign in with X and only for the actions you authorise — for example publishing a post you approved, reading your posts and mentions, or reporting ad-campaign performance; ad campaigns are drafted in paused state, and activating, pausing or re-budgeting a campaign always requires your explicit approval. The data exchanged with X in doing so is processed by X Corp. (USA) under the X privacy policy; we use data received from X’s APIs solely to perform these actions on your behalf and do not use it to train AI or machine-learning models.
• Web search (the web_search tool): AlphaAI Technologies Inc., trading as Tavily (USA), receives the search queries an agent formulates while working on your task — such a query can reflect the content of that task. Tavily’s standard terms allow it to use query data to improve its services and to pass queries to its own search and AI service providers; unless we have agreed data-processing terms with Tavily that exclude this, treat queries as shared with it on those terms, keep personal data, identification numbers and confidential details out of tasks that use web search, or remove the tool from the agent. Web search runs only for agents that have the tool; the starter team’s researcher has it from the start, and you can remove it from any agent. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Web search in an EU-hosted workspace (the web_search tool): Mistral AI (EU) runs the search and passes the search query an agent formulates to Brave Software, Inc. (USA), which operates the search service. A query can reflect the content of your task, and it leaves the EU although the workspace is EU-hosted. It is off where your workspace is restricted to the providers its employees are assigned to, and it runs only for agents that have the tool. Keep personal data, identification numbers and confidential details out of tasks that use web search, or take the tool away from the agent. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Embeddings (knowledge-base search): OpenAI (USA) computes text embeddings of the documents you upload — only where your data-residency policy permits US processing. Otherwise, in particular in an EU-hosted workspace, Mistral AI (EU) computes them; a local embedder is used only where Mistral is not configured. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Video generation (the optional generate_video tool): where the Seedance engine is configured on our deployment, BytePlus (a ByteDance company) receives the video prompt an agent writes and processes it in Malaysia (Johor) and/or Indonesia (Jakarta). Seedance counts as a China-region vendor, so it is reachable only if you have enabled the China region in your data-residency policy (off by default); otherwise a video is generated on Google Veo or HeyGen (both USA), or not at all. (no adequacy, no SCC — only with your Art. 17 DSG basis)
• Video generation (the optional generate_video tool): where the HeyGen engine is configured on our deployment, HeyGen Technology Inc. (USA) receives the video prompt an agent writes and processes it in the USA (on Amazon Web Services); your data-residency policy still gates each call. HeyGen’s privacy policy allows it to use what it receives to train its models unless the account has opted out. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Image generation (the optional FLUX engine of the generate_image tool and the “Create image” panel under Marketing): where FLUX is configured on our deployment, BFL GmbH (Black Forest Labs, Freiburg im Breisgau, Germany) receives the image description a person or an agent writes, and any of your own brand images chosen as a starting point for that image. We call its European endpoint, which processes in EU regions; your data-residency policy still gates each call. Black Forest Labs’ privacy policy allows it to use prompts, uploaded images and results to train its models unless the account has opted out. (adequate level of protection (Annex 1 DSV))
• Voice output (replies read aloud): where configured on our deployment, ElevenLabs, Inc. (USA) receives the text of a reply you have had read aloud, in order to synthesize the speech; your data-residency policy still gates each call. A reply can instead be voiced by Mistral (EU), Google (USA) or OpenAI (USA) from the list above, and every voiced reply names its vendor in the response. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Voice output: Cartesia AI, Inc. (USA) receives the same data for the same purpose and under the same conditions, where it is the speech vendor configured on our deployment. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Voice output: HeyGen Technology Inc. (USA) receives the same data for the same purpose and under the same conditions, where HeyGen Voice is configured on our deployment, and processes it in the USA (on Amazon Web Services). HeyGen’s privacy policy allows it to use what it receives to train its models unless the account has opted out. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Voice output, by choice only: Fish Audio receives the text of a reply you have had read aloud, in order to synthesize the speech, where it is configured on our deployment and an owner of your workspace has chosen it as the workspace’s voice. Fish Audio does not publish where it processes data, so we cannot name the state; no safeguard for a disclosure abroad is known to be in place, and the disclosure rests on your own basis under Art. 17 DSG. It is never used automatically and is offered only where your data-residency policy allows every region. (no adequacy, no SCC — only with your Art. 17 DSG basis)
• Video conversation (the optional face-to-face feature): where configured on our deployment, Tavus Inc. (USA) hosts the live video conversation you open with an AI assistant and receives your microphone audio and, if you switch it on, your camera video for the length of the conversation. We send nothing from your workspace, we do not record the conversation, and the AI assistant tells you at the start and on screen that it is an AI. Tavus’ own terms allow it to keep the conversation and use it to improve its service. The room is shown from an address of the real-time video provider Daily, which Tavus uses (section 9). Your data-residency policy and provider scope still gate the feature, and it processes nothing for a workspace that does not open a conversation. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Password check (self-service sign-up, password reset, and changing your password): Have I Been Pwned (operated by Troy Hunt’s Superlative Enterprises Pty Ltd, Australia) receives the first five hexadecimal characters of a SHA-1 of the password you chose, via the k-anonymity range API, so we can refuse a password that already appears in a public breach. Those five characters are shared by hundreds of thousands of possible passwords and cannot be related to you — they are not personal data. The password itself never leaves our server. A local denylist of common passwords still applies if that service is unreachable. (no personal data sent)
• Hosting: Render Services, Inc. (USA) runs the application in its Frankfurt region (Germany); its management systems and log stream are operated in the USA. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Database: the PostgreSQL database is provided by Neon (Databricks, Inc., USA) on Amazon Web Services in Frankfurt (Germany, region eu-central-1); the Neon platform uses Grafana Labs (USA) for infrastructure services. (adequate state; SCC for access from the provider's US group)
• Object storage (the files your agents produce, the documents filed in your workspace, and the nightly database backups described under “Retention”): Cloudflare R2 (Cloudflare, Inc., USA), in the storage region chosen when the storage is created. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Network edge and bot check: Cloudflare, Inc. (USA) fronts the application as CDN and reverse proxy. All traffic between your browser and Entumo — including the content you send and receive — passes, decrypted for the moment of transmission, through the nearest data centre of its worldwide network; Cloudflare does not keep your workspace content and logs request metadata (IP address, requested URL). Where the bot check is enabled on our deployment, opening the sign-up, sign-in or password-reset page loads Cloudflare’s Turnstile check, which receives your IP address and the browser signals it needs to tell people from bots. How Cloudflare processes that data is described in the Cloudflare Turnstile Privacy Addendum. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Email delivery (account e-mails — verification, invitations, password resets — and the notifications you configure, e.g. approval or run alerts, which can quote a short summary an agent wrote): Plus Five Five, Inc., trading as Resend (USA), receives the recipient’s email address and the message content — including the e-mails to our own inbox that report an access request, a bug report or an abuse report, which can contain the reporter’s name, e-mail address and text. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Payment processing (our own billing, independent of any connector you add): Stripe Payments Europe, Ltd. (Ireland) and Stripe, Inc. (USA) process the account holder’s billing details and payment method for subscriptions and credit top-ups. (FDPIC-recognised SCC; Swiss-US DPF where certified)
• Error monitoring: where error tracking is enabled on our deployment, Sentry (Functional Software, Inc., USA; data stored in its EU region in Germany) receives error reports from our server and from your browser — error type, message and stack trace, with pseudonymous user and workspace identifiers and the page address without its query part. Reports are scrubbed of credentials and never include your documents; an error message can occasionally quote a short fragment of the data being processed, for example an e-mail address. (adequate state; SCC for access from the provider's US group)
• OpenAI as sign-in provider (not a processor): only if you choose Sign in with ChatGPT, you sign in directly at OpenAI (USA), which thereby learns that you use Entumo and sends us the details named in section 2. OpenAI handles that sign-in as an independent controller under its own privacy policy; we send it nothing from your workspace.
• Public Swiss registers (not processors): when an agent looks up a company or searches the gazette, the federal UID register (Federal Statistical Office) and the Swiss Official Gazette of Commerce (SECO) receive the name or UID looked up — for a sole proprietorship that can be a person’s name; an exchange-rate lookup (Federal Office for Customs and Border Security) receives only a date.
• Partners: a partner who referred your workspace (see section 2) sees its name and creation date.

Version and history of this list. Current version: 2026-10-18.

• 2026-10-18: Added NVIDIA Corporation (NVIDIA NIM, USA) as an optional model provider (Nemotron 3.5 Super VL 120B); off unless configured and the workspace's data-residency policy allows the USA, and for workspaces created before 18 October 2026 it starts on 17 November 2026, or earlier if an owner accepts this list. In addition, the roles of OpenAI and Google now name every use the privacy policy already describes: besides running agent models, OpenAI transcribes dictation, reads replies aloud and generates images, and Google transcribes dictation, generates images and runs the Veo video engine — each only where configured and where the workspace's data-residency policy allows the USA.
• 2026-10-17: Added BFL GmbH (Black Forest Labs, Freiburg im Breisgau, Germany) as an optional image generation engine (FLUX). It receives the image description a person or an agent writes, and any of the workspace's own brand images chosen as a reference for that image. We call its European endpoint, which processes in EU regions. Black Forest Labs' privacy policy allows it to use prompts, uploaded images and results to train its models unless the account has opted out. It runs only where configured on our deployment. For workspaces created before 17 October 2026 it starts on 16 November 2026, or earlier if an owner accepts this list.
• 2026-10-16: Added HeyGen Technology Inc. (USA) for two optional features: HeyGen Voice as a voice for replies read aloud — it receives the text of a reply you have read aloud, to synthesize the speech — and HeyGen Video as an engine of the video generation tool — it receives the video prompt an agent writes. HeyGen processes in the USA (on Amazon Web Services); the disclosure rests on the Standard Contractual Clauses and the Swiss-US Data Privacy Framework. HeyGen's privacy policy allows it to use what it receives to train its models unless the account has opted out. Both run only where configured on our deployment and only where the workspace's data-residency policy allows the USA. For workspaces created before 16 October 2026 they start on 15 November 2026, or earlier if an owner accepts this list.
• 2026-10-14: Added Fish Audio (Drama 3 voice model) as an optional voice for replies read aloud. It receives the text of a reply you have read aloud, to synthesize the speech. Fish Audio does not publish where it processes data, so its location is listed as unknown and no transfer safeguard is known to be in place. It is never used automatically: only if an owner chooses it as the workspace's voice, and only where the workspace's data-residency policy allows every region. For workspaces created before 14 October 2026 it can be chosen from 13 November 2026, or earlier if an owner accepts this list.
• 2026-10-07: Web search for EU-hosted workspaces: Mistral AI (EU), already listed, runs the search and passes the search query to Brave Software, Inc. (USA). A search query therefore leaves the EU even in an EU-hosted workspace. It is off where a workspace is restricted to its assigned providers. For workspaces created before 7 October 2026 it starts on 6 November 2026, or earlier if an owner accepts this list.
• 2026-10-06: Mistral AI (EU), already listed, now also computes the text embeddings for knowledge-base search wherever OpenAI's embedder may not run — in particular in EU-hosted workspaces, which used a local embedder until now — and is the default model provider for EU-hosted workspaces. For workspaces created before 6 October 2026, Mistral starts computing embeddings on 5 November 2026, or earlier if an owner accepts this list. No processor added; no processing location or transfer basis changed.
• 2026-10-04: Added Xiaomi MiMo as an optional model provider in the China policy region. Off unless configured and the workspace permits that region; existing workspaces require the notice period or early owner acceptance.
• 2026-10-03: Updated the default manager model name from GPT-6 Sol to GPT-6.1 Sol. No processor, processing location or transfer basis added or changed.
• 2026-10-02: Added Tavus Inc. (USA) for the optional face-to-face feature: a live video conversation a person opens with an AI assistant. While the conversation runs, Tavus receives the person's microphone audio and, if they switch it on, their camera video, and hosts the conversation room. Nothing from the workspace is sent. The feature is off unless we configure it, only available to workspaces whose data-residency policy allows the USA and whose provider scope is not restricted to assigned providers, and always tells the person they are talking to an AI. Tavus' terms allow it to keep the conversation and use it to improve its service. It processes nothing for a workspace that does not open a conversation.
• 2026-09-30: Corrections; no processor added or replaced. Locations now state where processing can actually run: Anthropic and Google in the USA or worldwide (Anthropic routes requests globally by default; Google may log or cache transiently in any country it operates in); Alibaba Cloud with storage in Singapore and processing on international nodes outside mainland China; BytePlus in Malaysia and/or Indonesia; Render in Germany for the application, with its management systems and logs in the USA; Cloudflare R2 is no longer described as EU jurisdiction (the region is chosen when the storage is created). Contracting entities are named: Anthropic Ireland, OpenAI Ireland, Databricks for Neon, Plus Five Five for Resend and AlphaAI Technologies for Tavily. OpenAI is described as the default manager model, Tavily's terms on the use of queries are stated, and Resend's scope includes the e-mails to our own inbox about access, bug and abuse reports.
• 2026-09-23: Added two EU-resident model providers for EU-hosted workspaces: OpenAI's European data-residency region (eu.api.openai.com — OpenAI is already listed; this is its separate EU processing region, where prompts are processed and stored in Europe) and Google Cloud Vertex AI's EU multi-region (Gemini and Claude Fable, with ML processing inside EU member states — Google is already listed for the Gemini API). Neither processes anything for a workspace that does not use its models. Same revision: every entry now names the state it processes in and the transfer basis, and the platform processors (Render, Neon, Cloudflare R2 and edge, Resend, Stripe, Sentry) — already named in the privacy policy — are part of the versioned list. Two locations were corrected, not changed: Have I Been Pwned is operated from Australia (it receives a five-character hash prefix, not personal data) and the Seedance video engine processes in Johor, Malaysia.
• 2026-08-13: Added Have I Been Pwned (Troy Hunt). On self-service signup, password reset and in-app password change we send the first five hex characters of a SHA-1 of the chosen password to the k-anonymity range API so a password that already appears in a public breach is refused. The password itself never leaves the server.
• 2026-08-07: First versioned publication. No processor added or removed — the inventory below is the one already in force; this revision only puts a version and a change history on it.

We do not sell personal data. Each processor engages its own sub-processors under its own terms; they appear in the processors’ published sub-processor lists. We disclose personal data to authorities only on the basis of a Swiss order or a statutory duty, and we check each request. We do not release data directly to a foreign authority or court without a procedure recognised in Switzerland (mutual legal assistance; Art. 271 and 273 of the Swiss Criminal Code), and we refer such requests to that route. Where a request concerns your workspace we inform you first, unless the law or the authority’s order prohibits it (terms, section 12(a)). This does not bind our processors, which are subject to the law of their own states and may receive orders directly (see section 5).

5. Disclosure abroad

Our application and database run in Frankfurt (Germany), a state with an adequate level of data protection (Annex 1 DSV); our files and backups are stored with Cloudflare R2 in the region named in section 4. Several processors belong to US groups, and some process data in the USA or worldwide — in particular Anthropic (Claude), which powers the default team and, because the provider may route requests anywhere in its network, can process them in the USA or elsewhere in the world, OpenAI, which runs the manager by default, Cloudflare’s network and, for the optional face-to-face conversation, Tavus. For those recipients, and for access from the USA, we rely on the EU Standard Contractual Clauses in the version recognised by the Swiss Federal Data Protection and Information Commissioner (with the Swiss amendments) and, for certified recipients, the Swiss-US Data Privacy Framework. The state and basis for every processor are listed in section 4. Anthropic is contractually bound not to use your content to train its models. Model providers may keep inputs and outputs for limited periods under their own terms — for example for abuse monitoring and safety review, and longer where their safety systems flag content — regardless of how long we keep them. Your data-residency policy on the Security page controls which jurisdictions may process your data (EU/US/CN) and is enforced fail-closed — it defaults to the EU and the USA, so no data reaches a China-region provider unless you explicitly enable the China region. The optional China-region providers (Moonshot/Kimi and Z.AI in China; Alibaba Cloud/Qwen with storage in Singapore and processing on international nodes worldwide outside mainland China) are used under their own standard API terms; we have not concluded Standard Contractual Clauses with them and their terms may permit them to use content to improve their services; the same applies to the Seedance video engine of BytePlus (Malaysia and/or Indonesia), where it is configured on our deployment, and to Fish Audio, which does not publish where it processes data and voices a reply only if an owner chooses it and every region, including China, is enabled. Enabling the China region and assigning work to those models is a disclosure abroad to states without an adequacy decision and without additional safeguards — do so only for data for which you, as controller, have a lawful basis under Art. 17 revDSG (in particular the express consent of the persons concerned). If the GDPR applies to your processing, do not enable the China region for personal data: no transfer tool and no Art. 28(4) GDPR terms exist for those providers. A provider that does not offer even the contractual data-protection commitments we require of a sub-processor is not offered at all: DeepSeek, for example, appears as “coming soon” and cannot be selected or enabled.

EU hosting. If you choose EU hosting when you create a workspace — or leave only the EU region ticked on the Security page — your AI processing uses only the European zones of the providers: AWS Bedrock EU (Frankfurt, Zurich, Stockholm, Milan, Spain, Ireland, Paris), Microsoft Azure’s EU Data Zone (which may include Switzerland and Norway), OpenAI’s European region (EEA and Switzerland), Google Cloud Vertex AI EU and Mistral — all in states with an adequate level of protection. Except for Mistral these zones are operated by companies of US groups, so access from the USA cannot be ruled out; the safeguards above cover it. Features with no European provider are unavailable in an EU-hosted workspace: image and video generation, the face-to-face video conversation and US voice vendors. Web search runs through Mistral AI (EU), which passes the search query to Brave Software, Inc. (USA), so a search query does leave the EU; it is off where the workspace is restricted to its assigned providers. Knowledge-base embeddings are computed by Mistral AI (EU) instead of OpenAI, and Mistral is the default model provider. EU hosting does not change the platform processors — Cloudflare’s network, Stripe, Resend and Sentry, Render and Neon — nor the connectors you choose. It is offered only where our deployment has a European provider configured.

6. AI-assisted processing and automated decisions

Entumo uses AI agents that read and act on data automatically; you are interacting with an AI system and the work it produces is AI-generated, which can be incomplete or incorrect — review it before you rely on it. How it works: an employee receives your brief together with the context it needs — your company profile, the relevant passages of your knowledge base, its memory, and what it reads in the tools you connected or on the web — sends it to the AI model that runs it, and proposes or carries out the actions you permitted, for the purpose of doing the work you assigned. Actions that carry real-world consequences — moving money, messaging a customer, writing a financial record, signing — are routed through a guardrail that pauses them for your explicit human approval before they run. That approval requirement is the default; a workspace owner can relax individual pauses or switch the guardrail to a monitor-only mode on the Security page (which warns about the implications) — the payment cap, the block on irreversible deletions and the credential-leak block remain enforced in every mode. Our own processing of your account takes no automated individual decision about you. Where your agents’ work could amount to an automated individual decision about a person outside your company (Art. 21 revDSG), your company is the controller and owes that person the information and a human review (section 1); the approval pause helps you provide it. Entumo does not rank, filter or reject job applicants and does not evaluate employees: its recruitment workflow summarises applications against the criteria you set and leaves every decision to you.

Transparency: Entumo is an AI system within the meaning of Regulation (EU) 2024/1689 (EU AI Act), and the product tells you when you interact with an AI agent — the work your agents produce is attributable to them as AI-generated. The face-to-face assistant is an AI as well: it says so when the conversation opens, a notice stays on your screen for the whole conversation, and it takes no decision about you. When your agents write to people outside your company, the platform itself places a notice at the start of the message stating that it was written by an AI agent on behalf of your company — on emails sent through a connected mailbox or e-mail service, postal letters rendered by the platform, messenger and SMS messages, helpdesk and review replies and the other outbound channels that carry text the agent wrote; at the end of a public social-media post — and marks those emails with a machine-readable header (Art. 50(1) and (2) EU AI Act, applicable since 2 August 2026; the Swiss FDPIC likewise holds that people are entitled to know whether they are corresponding with a machine). This notice cannot be switched off. Internal team channels, machine-to-machine endpoints and notifications a provider composes from its own template carry none; where a channel cannot carry it (for example a pre-approved WhatsApp template or a custom connector), the agent is instructed to state it. Images and videos your agents generate carry machine-readable metadata identifying them as AI-generated; a generated image or video that an agent posts to X is also sent with X’s own AI flag (made_with_ai) where X accepts it for our application, and otherwise labelled in the post text where it fits; and an agent refuses to depict a real, identifiable person unless you confirm that person’s consent. Entumo itself never trains models on your data. Your prompts and content are sent to the AI model providers described in section 4 solely to generate the response; what a provider may do with them is set by its API terms — Anthropic (Claude) does not use API content to train its models, while Moonshot AI’s (Kimi) standard terms let it use content to improve its services unless restricted by a separate written agreement. Mistral’s terms let it use content to train its models unless the customer has opted out in its account settings, and Tavily’s standard terms let it use queries as described in section 4. Tavus’ terms let it keep a face-to-face conversation and use it to improve its service (section 4); an owner who does not want that switches the feature off by restricting providers to those its employees are assigned to (Security page). Assign work whose content must not be used that way to Claude-powered employees and use the provider setting in section 4. The same applies to every other selectable provider — for example Alibaba Cloud (Qwen), Z.AI or, once offered, DeepSeek: unless this policy states a no-training commitment for a provider, assume its API terms allow such use. In Switzerland there is currently no AI-specific statute: AI-assisted processing of personal data is governed by the revDSG, and Switzerland has signed the Council of Europe Framework Convention on Artificial Intelligence, whose implementation the Federal Council intends to put to consultation — we track it.

7. Retention

We keep personal data only as long as needed for the purposes above. In practice: account data for as long as your account exists — you can delete your account yourself on the Security page, and it is then removed from the live system at once; because the records you created belong to your workspace, your name and email address are replaced there by a neutral placeholder (for example in the workspace audit trail). An account whose email address was never confirmed and that belongs to no workspace is deleted after 7 days. Workspace data is kept until the workspace is deleted — by its owner at any time, or by us 90 days after its paid subscription ended without renewal, unless prepaid credits remain, a payment dispute is open or a legal hold applies (you can export it until then); when your account or workspace is closed or you ask us to delete it, we delete it within 30 days — deleting a workspace removes its active agent, run and connection records from the live system immediately. File erasure and provider revocation continue through the cleanup process below. Request and application logs (IP address, timestamp, user-agent and, for account events, the e-mail address or user id) exist only in our hosting provider’s log stream and are deleted there automatically within at most 30 days — we keep no separate log archive. Sign-in sessions end after 12 hours without activity and at the latest after 48 hours; abuse counters within a day after their window; product statistics after 180 days; aggregate page counts after 365 days; access requests after 12 months. The workspace audit trail keeps the most recent 2,000 human actions and the most recent 2,000 agent actions and is deleted with the workspace. Where the law requires us to keep our own records longer — for example our accounting records under the Swiss Code of Obligations, such as the credit ledger and the invoices for your subscription — we keep those records for the statutory period (generally 10 years) and then delete them.

Your business records. The receipts, letters, contracts, year-end papers and other records your workspace keeps are your company’s records, not ours: we keep them until you delete them or the workspace, and they are deleted with the workspace. Entumo is not an archive under the Swiss accounting records ordinance (GeBüV). Your company must keep its books and vouchers for ten years (Art. 958f CO); keep them in your accounting software or another archive, and take the workspace export and the files before you delete the workspace or let a subscription lapse.

Records of your agreements. Each time you accept the terms — at sign-up, by invitation or through the notice shown after a new version — we keep a record: the version of the terms and of this policy, a fingerprint (SHA-256) of their text, the language you read them in, how you accepted and the time; we record no IP address. We keep the same kind of record of an owner’s agreement to automatic top-up and to the China data region (the owner, the time, the wording and its fingerprint), with the workspace’s wallet and data-residency settings. If you delete your account, the acceptance records remain without your name, e-mail address or user id, under a keyed pseudonym of the address, for 10 years, so that we can show what was agreed if a claim is made (limitation period of Art. 127 CO); your acceptances are included in the export of your account.

Face-to-face conversations. We do not record a face-to-face conversation and keep no audio, video or transcript of it. The room closes when you end the conversation, when the time limit shown before you start is reached, or shortly after you leave. We keep only the usage and credit entry for it — when it took place and what it cost — like any other charge on your account. How long Tavus keeps what it received is set by its own terms (section 4); we do not set that period.

Deletion cleanup. To finish erasure and withdraw connector access reliably, we retain a restricted cleanup record with workspace/run identifiers, encrypted credential snapshots and the encrypted storage configuration needed for that task. Successful revocation removes the corresponding credentials from the cleanup record; failed erasure or revocation is retried and can require operator intervention. These records are not an active workspace and are not used to run new business tasks. Minimal erasure markers and manual-provider follow-up metadata may remain without credentials or document content. Local deletion does not prove that a provider withdrew access or deleted its own copies. You may need to revoke access directly in that provider’s account. Cleanup remains subject to the purpose and deletion obligations above; retrying is not permission to retain data indefinitely.

Backups. So that we can restore the service after a loss or corruption incident, a complete copy of our database is written every night to our object-storage provider (Cloudflare R2). It covers every table — including account data, workspace content and the encrypted connector credentials — and the 30 most recent nightly copies are kept, each older one being deleted automatically. Deletion of active workspace records is immediate as described above, and we restore from a backup only to recover the service as a whole, never to bring individual deleted records back; a copy made before a deletion does, however, still contain the data until it ages out of that 30-night window.

Payment records. When a workspace is deleted — by you, by us after a lapsed subscription, or together with your account — its subscription is cancelled and we ask Stripe, our payment processor, to delete the workspace’s customer record: name, e-mail address, billing details and saved payment method. This runs in the background after the deletion and is repeated until Stripe confirms it; if a payment dispute is open at that moment, we keep the customer record until a person has dealt with it, and where it cannot be done automatically a person completes it. Stripe keeps the records of the charges and invoices themselves under its own retention rules and legal duties, and we keep the credit ledger and invoices for the statutory period (generally 10 years). A saved card can also be removed earlier, while the workspace exists, on written request.

In addition to cleanup and backups, some records outlive workspace deletion: our own accounting records above and an abuse report concerning your workspace: we keep the report together with the evidence snapshot taken when it was filed (the reported content, its checksum, the agent that produced it and the related audit entries) for as long as we need it to handle the case and to establish, exercise or defend legal claims. The party a report is about controls every delete surface in the product, so evidence that vanished with the workspace would make deleting the workspace the universal way to erase it. Reports that are still open or escalated are kept while the case is open. A report found unfounded is deleted, with its evidence, 30 days after its last update; a report that led to action or was otherwise closed, 2 years after its last update. The reporter’s name and contact details are removed from a decided report 30 days after its last update, and earlier on request unless a case is pending. Reports concerning child sexual abuse material are not deleted automatically, because criminal law decides how they are handled. A legal hold on the workspace concerned suspends deletion.

Bug reports. When you report a problem with the product from inside the app, we store what you wrote, how urgent you said it is, your name, e-mail address and workspace, and — only if you leave the box ticked — the page you were on (its web address without the query part, so no one-time link travels with it), your language, your browser, your screen size and the last few errors your browser had already logged. We use it to reproduce and fix the problem and to reply to you. It is deleted with your workspace. A report filed before you belong to any workspace is kept under our own platform record instead, where the 500 most recent are retained and older ones are deleted automatically.

8. Security

Connector credentials are encrypted at rest (AES-256-GCM, with a key derived per workspace) and decrypted only server-side at the moment a connector runs; they are never returned to the browser. Passwords are stored only as salted hashes and session identifiers only as digests; two-factor sign-in is required for every account by default, and sensitive actions require you to re-enter your password. The database and file storage are encrypted at rest by their providers, access is scoped per workspace, and transport is over HTTPS. Every consequential agent action passes the approval guardrail and is recorded in the workspace audit trail. The full list of technical and organisational measures is the annex to section 12 of the terms.

9. Cookies and storage in your browser

We use only strictly necessary first-party cookies and no advertising or tracking cookies, so under Swiss law (Art. 45c FMG) no consent banner is required:

• vorenq_session — keeps you signed in; ends after 12 hours without activity, at the latest after 48 hours.
• vorenq_stepup — confirms for 15 minutes that you re-entered your password before a sensitive action.
• vorenq_company — which workspace you last opened; kept for 48 hours (as long as a sign-in can last) and removed when you sign out.
• vorenq_terms_ack — which version of our terms you have acknowledged; one year.
• NEXT_LOCALE — your language: until you close the browser, or one year when you choose the language yourself in the language menu. It is set when you choose a language, or when the page language differs from the language preselected for you; that preselection comes from the country our network provider derives from your IP address, which is not stored, or from your browser’s language.
• Short-lived state cookies (10 minutes) that secure a connector sign-in, or a sign-in with ChatGPT (oauth_chatgpt_login).
• Our network provider (Cloudflare) can set a strictly necessary security cookie when it screens automated traffic.

The cookies we set carry the Secure attribute, so your browser sends them only over an encrypted connection. In your browser’s local storage we keep your light/dark theme, interface preferences (which chats you have read, recently used commands, view modes, dismissed hints, your dictation shortcut) and unsent drafts of chat messages — the drafts are deleted when you sign out. Session storage, which ends with the tab, holds a temporary copy of workspace data for faster navigation (also cleared when you sign out or switch workspace), a draft of the onboarding form and the marketing context described in section 2. None of this is shared with third parties. You can block or delete cookies and site data in your browser; without the session cookie you cannot sign in, while the public pages keep working. This assessment follows Swiss law; if you visit from the EU/EEA, note that the marketing context in session storage (section 2) is not needed for the site to work, and you can refuse it at any time by blocking site data for this domain.

During a face-to-face conversation the video room is shown from the address tavus.daily.co, operated by the real-time video provider Daily on behalf of Tavus. Your browser connects to it directly, so it receives your IP address and technical data, and it may keep its own cookies or local storage on your device for the call — we do not set or read them. Nothing is loaded from that address until you press Start.

10. Your rights

You have the right to information about your data — free of charge, answered within 30 days —, to receive your data in a commonly used electronic format (data portability), and to its correction, deletion, restriction of processing, and to object to processing. Much of this you can do yourself: download an export of your personal data and delete your account on the Security page, correct your name and email address there, and — as a workspace owner — export the workspace on the Data control page. Otherwise contact us at info@momentumq.com or by post at the address in section 1; we may ask you to confirm your identity. For data that a company processes about you with Entumo, contact that company (section 1). Data that Tavus holds about a face-to-face conversation is not stored by us: write to us at the address above and we will pass your request on to Tavus. You may lodge a complaint with the Swiss Federal Data Protection and Information Commissioner (FDPIC/EDÖB); if you are in the EU/EEA, you may also complain to your national data protection authority.

Children. Entumo is a service for organisations and is not directed at children; a person must be at least 15 years old to use it (terms, section 2). If you believe that personal data of a child has been entered into Entumo, write to us at the address above; we delete it unless the law requires us to keep it.

11. Changes

We may update this policy. It informs you about our processing; it is not part of the terms, and updating it does not amend them — changes to the parts of it that section 12 of the terms refers to (the sub-processor list, supporting tasks, deletion and retention) follow section 12(d) and section 14 of the terms. Before a new sub-processor starts processing the data of an existing workspace, we notify its owners by email and on the in-app Security page at least 30 days in advance, stating the date the change takes effect; a newly added AI model provider does not process your workspace’s data before that date unless an owner accepts the updated list earlier (terms, section 12(d), which also sets out your right to object). Last updated: 18 October 2026. Each version of this policy is listed below, and earlier versions are available on request.

Version history. 2026-10-18: Holidays and absences are now deleted automatically five years after the end of the year in which they ended (wage claims can be raised for five years, Art. 128 CO), except while a legal hold applies; open requests stay until they are decided (section 2, data used by optional workflows). Previously your company decided how long they were kept.