Legal

Sub-processors

Last reviewed: 14 October 2026

David is a data processor acting on your instructions; you are the controller. The parties below are the sub-processors Ergonomy Limited engages to provide the Service. This list is referenced by our Privacy Policy and Data Processing Agreement, and the technical detail (what an AI model sees, retention, residency) sits on our data-residency page.

How to read this list

Your administrators choose in David's settings whether AI model requests stay in the UK, stay in the EU, or may use all providers. Each vendor below is marked with how it is engaged for your organisation:

Admin-enabled An AI model provider. Engaged only for the models your administrators enable on the Models page, within the AI processing region they choose. Each row says whether the standard configuration enables any of its models.
Default A supporting vendor engaged in the standard configuration. It can be excluded for your organisation on request, and we tell you which features, if any, switch off as a result.
Feature-gated Receives data only while you use the named feature (for example, recording a meeting or connecting a specific tool).

To keep AI model requests inside the UK or EU, use the region setting in David; to enable or remove a model provider, use the Models page. To exclude a supporting vendor, email legal@meetdavid.ai (see clause 5 of the Data Processing Agreement).

AI model providers

Which of these providers is engaged follows two choices your administrators make in David’s settings: the AI processing region, and which models users may run (the Models page). A provider whose models are not enabled is not engaged for model requests. The standard configuration enables models from the providers marked accordingly below. Credentials and integration secrets are never sent to a model provider.

Sub-processor	Purpose	Engagement	Data location	Entity & jurisdiction
Amazon Web Services (AWS Bedrock)	Claude and open-weight model inference run through AWS	Admin-enabled In the standard configuration.	Region-pinnable: UK (London), EU, or US	Amazon Web Services, Inc. (US; UK and EU regions available)
Google (Vertex AI)	Gemini model inference and image generation	Admin-enabled In the standard configuration.	Region-pinnable; EU broadly, London (UK) for a limited set	Google LLC (US)
Anthropic	Claude model inference (first-party API)	Admin-enabled In the standard configuration.	US; EEA/UK contracting via Anthropic Ireland Ltd	Anthropic PBC (US) / Anthropic Ireland Ltd (EEA/UK)
OpenAI	GPT model inference (business/API tier), text-to-speech for spoken replies, and image generation	Admin-enabled In the standard configuration.	US default; EU residency for supported endpoints	OpenAI, L.L.C. (US) / OpenAI Ireland Ltd (EEA)
Groq	Fast inference of open-weight models	Admin-enabled In the standard configuration.	US	Groq, LLC (US)
Fireworks AI	Inference of open-weight models	Admin-enabled In the standard configuration.	US (EU via enterprise dedicated deployments)	Fireworks.ai, Inc. (US, Delaware)
Meta Platforms	Muse Spark model inference (standard tier only, which does not train on prompts)	Admin-enabled Not in the standard configuration.	US	Meta Platforms, Inc. (US)
xAI	Grok model inference (developer API)	Admin-enabled Not in the standard configuration.	US default; EU endpoint by agreement	xAI (US)
Telnyx Inference	Open-weight model inference on Telnyx GPUs. Receives prompts, selected conversation and document context, tool results, supported images, and generated replies. Data processing terms AI service terms Retention and residency Inference service	Admin-enabled Requires organisation-level enablement by Ergonomy and an enabled model, including where an automatic model route selects Telnyx. Unavailable under David’s UK or EU AI-processing setting. Not in the standard configuration.	US-preferred routing; processing may move to other regions. No UK/EU pin configured in David.	Telnyx LLC (United States)

Search, classification, speech, and observability

Engaged for search and retrieval, for quick classification decisions, for specific features (meeting transcription, dictation), and for service observability. Feature-gated rows receive data only while you use that feature.

Sub-processor	Purpose	Engagement	Data location	Entity & jurisdiction
Voyage AI	Text embeddings so David can search your conversations and knowledge	Default	US	Voyage AI, a MongoDB company (US)
TypeSafe	Quick classification decisions, such as how much reasoning a request needs, whether a message in a shared workspace is addressed to David, whether an automation’s condition is met, or the next step of a browser task. Receives the text each decision is about (a message and recent conversation, an incoming event, page text, or an item in a batch job) and returns a choice or a score, not generated text. Data processing terms Privacy policy	Default In the standard configuration for new organisations. An organisation set up before then uses it only where it has been switched on for that organisation.	US, under every AI processing setting	TypeSafe AI, Inc. (US)
Tavily	Web search on queries derived from your request	Default	US	AlphaAI Technologies Inc. d/b/a Tavily (US)
Exa	Web search and retrieval on queries derived from your request	Feature-gated Only where Exa is configured as the web-search provider for your organisation.	US	Exa (US)
Deepgram	Speech-to-text for recorded meetings	Feature-gated Only when you record a meeting with the desktop app.	US	Deepgram, Inc. (US)
Gladia	Speech-to-text for recorded multi-speaker meetings	Feature-gated Only when you record a meeting with the desktop app.	EU (France)	Gladia SAS (France)
AssemblyAI	Live dictation: audio streamed from your browser while you dictate a message	Feature-gated Only while you use dictation in the message box.	US	AssemblyAI, Inc. (US)
Langfuse	Observability and tracing of AI requests, used for reliability and support investigation; traces include prompts and outputs	Default	EU (Ireland, AWS eu-west-1)	Langfuse (ClickHouse, Inc., US); EU data region

Integration brokers

Holds the authorised connection to a third-party tool on our behalf and relays requests to it. Most integrations run through our own self-hosted broker, which is not a third party.

Sub-processor	Purpose	Engagement	Data location	Entity & jurisdiction
Composio	Integration broker for some connected tools: holds the authorised connection and relays requests to the tool	Feature-gated Only when you connect a tool through a Composio-backed integration.	US	Sampark, Inc. d/b/a Composio (US, Delaware)

Billing, infrastructure, and hosting

Compute, storage, and hosting that underpin the Service, and payment collection for paid plans.

Sub-processor	Purpose	Engagement	Data location	Entity & jurisdiction
Amazon Web Services (AWS)	Compute, object storage (KMS-encrypted), queuing, backups, hosting, and Polly text-to-speech where selected	Default	Per-tenant region (UK / EU / US)	Amazon Web Services, Inc. (US; UK and EU regions available)
Stripe	Payment collection (Direct Debit and card), mandates, invoices, and the billing contact details needed for them	Feature-gated Only when you take a paid plan or top up credits.	Contracting in Ireland; processing may occur in the US under SCCs and the UK Addendum	Stripe Payments Europe, Limited (Ireland) for UK and EEA customers
Google (Web Push / FCM)	Transport for encrypted browser notifications	Feature-gated Only when browser notifications are configured and you enable them. Your browser selects the transport.	Browser-selected service; no regional routing guarantee	Google LLC (US)
Mozilla (Autopush)	Transport for encrypted browser notifications	Feature-gated Only when browser notifications are configured and you enable them. Your browser selects the transport.	Browser-selected service; no regional routing guarantee	Mozilla Corporation (US)
Apple (APNs)	Transport for encrypted browser notifications	Feature-gated Only when browser notifications are configured and you enable them. Your browser selects the transport.	Browser-selected service; no regional routing guarantee	Apple Inc. (US)
Microsoft (WNS)	Transport for encrypted browser notifications	Feature-gated Only when browser notifications are configured and you enable them. Your browser selects the transport.	Browser-selected service; no regional routing guarantee	Microsoft Corporation (US)

Not sub-processors

For the avoidance of doubt, these are not sub-processors of Ergonomy:

Your connected tools and channels (for example Microsoft 365, Google Workspace, Xero, Microsoft Teams, WhatsApp). David accesses them with the permissions you grant, and they process data under your own arrangements with them.
Sign-in providers. Signing in with Microsoft or Google sends David only the identity assertion (identifier, verified email, name, avatar); no content flows to the provider through login.
Support partners you authorise through the Service act under your instructions, not ours.
Our own infrastructure: the vector index, the integration broker for most connected tools, and speaker recognition for meetings all run inside your environment on our infrastructure. No third party receives that data.

Change log

14 October 2026

Added TypeSafe, which makes quick classification decisions inside David and returns a choice or a score, not generated text. It processes in the US under every AI processing setting and is in the standard configuration for new organisations.
12 October 2026

Added Telnyx Inference as an optional model provider, with its processing location, retention and DPA. It is not enabled in the standard configuration.
7 September 2026

Named the feature and platform vendors previously described only by category (Deepgram, Gladia, AssemblyAI, Voyage AI, Tavily, Exa, Composio, Langfuse). Added Stripe (billing) and the model providers Meta Platforms and xAI, which the standard configuration does not enable. Removed Nango: it now runs on our own infrastructure, so it is no longer a third party. Marked every vendor as admin-enabled, default, or feature-gated.
9 July 2026

First published list.

Notification of changes

We give each affected customer's designated account or privacy contact at least 30 days' written notice before a new or replacement sub-processor receives their personal data. The change log supplements that notice; no separate subscription is required. You may object on reasonable data-protection grounds during that period, and we do not use a disputed sub-processor for your data while the objection is being resolved. See clauses 5.6–5.7 of the Data Processing Agreement.