AI Notice

This notice explains how we use artificial intelligence (AI) across PartnerAPI in a way that satisfies the transparency obligations of Article 50 of the EU AI Act (Regulation (EU) 2024/1689). It complements Section 7 of our Privacy Policy, which describes how personal data is processed by these features.

Where AI is used in PartnerAPI

We use AI in the following user-facing surfaces:

• Partner assistant — an in-product chat assistant that helps partners explore the API, summarise documentation, surface integration diagnostics, and draft integration steps.
• Tenant-admin assistant — an in-product chat assistant for tenant administrators that helps pull partner metrics, draft summaries, and suggest next steps.
• Action-card copy — the short title and text partners see on an action card can be written by AI from the card's configuration; if generation fails, the tenant administrator's own wording is shown instead.
• Partner and prospect discovery recommendations — where a tenant administrator runs a discovery or partnership review, candidate companies are ranked by a deterministic, non-AI score, and the assistant then writes the recommendation on each one: a verdict (approve, needs more info, not a fit), a confidence, the reasons and the supporting evidence. Those recommendations are model-authored and are labelled as AI-generated where they appear. The assistant decides nothing — a person takes every decision, and no partner status changes without their explicit action.
• AI-drafted replies — the assistant can draft replies to partner conversations held by email or Slack. A person at the business using PartnerAPI reviews and approves every reply before it is sent. That business can choose to end each reply sent this way, by email, in Slack or on a support ticket, with a line saying it was written with help from AI and naming the person who reviewed it. The line is off unless the business turns it on.
• Retrieval-augmented documentation — uploaded documentation, API documentation and support tickets are processed for embedding-based search so the assistants can cite the right source on request. Uploaded documents are first screened by an automated moderation model, and a document it flags is not indexed.
• AI-generated landing pages — tenant administrators can have their partner landing page generated or revised from a written description and, optionally, a reference page. Generated pages carry a visible note and a machine-readable marker saying they were generated with AI assistance.
• Agent and certification drafts — tenant administrators can generate draft agents (troubleshooting scenarios with steps and resolutions) and draft certification templates, with suggested API endpoints, from their own API documentation. These are drafts for an administrator to review and edit.
• Upstream API change summaries — when a third-party API that an integration depends on changes, a one-sentence plain-language summary of the change is written by AI. The list of changes and their severity are determined without AI.

An inventory of the AI systems behind these features is maintained internally, and material information about it is available on request — see the contact details in our Privacy Policy.

What the AI can and cannot do

The AI features above can:

• Summarise text, draft messages, suggest next steps, and answer questions from documentation you provided.
• Produce first drafts — landing pages, agents and certification templates — for an administrator to review.
• Surface diagnostics and integration hints in the assistant chat surfaces.

The AI features above cannot:

• Take destructive or irreversible actions on your behalf without your explicit confirmation.
• Replace human review on decisions that affect partner status, billing, or compliance.
• Guarantee accuracy. Always verify anything material before acting on it.

When you interact with an AI assistant we tell you so on the first message of the conversation, and we keep that disclosure visible in the transcript. The disclosure is mandatory and cannot be disabled by a tenant.

Models and providers

The AI features above rely on these third-party providers, each listed under AI & Machine Learning on our subprocessor list:

• OpenAI — the large language models behind the assistants and every other feature above except landing pages, the embeddings used for document search, and the moderation of uploaded documents.
• Vercel (v0) — generation of tenant landing pages.
• Chroma Inc. (Chroma Cloud) — storage of the embeddings used for document search.

The authoritative, current list of subprocessors — including every model provider — is maintained at /legal/subprocessors. Changes to that list are published on that page and notified to customers as set out in our Data Processing Agreement.

How your data is used

Personal data handling for AI features is described in Section 7 of our Privacy Policy. In summary:

• Prompts and outputs sent to OpenAI are processed strictly to generate a response; they are not used to train OpenAI's models.
• Conversation history is retained for the durations described in our Privacy Policy.
• Tenants can disable specific AI features at the tenant level (assistant, document embedding, AI data-processing opt-out, conversation logging) via the tenant settings surface. Per-user opt-out for end users where legally available is in development.

Your rights

You can:

• Ask which AI features apply to you — write to us using the contact details in our Privacy Policy.
• Ask why you saw something — consequential AI outputs carry a “Why this?” link that names the AI system behind the output and what it was given to work from.
• Request human review of a consequential AI-assisted decision — the same panel offers “This looks wrong”. A person reviews these reports and writes back with an explanation; the outcome is recorded against the original decision and shown to you on your account's Privacy & AI page.
• Object to or opt out of processing as permitted by GDPR — see Section 8 of our Privacy Policy.

If you believe an AI feature has produced an unlawful outcome, you may also lodge a complaint with your national supervisory authority for the AI Act in your EU member state.

Synthetic content

Where this product generates content that could be mistaken for human-authored or human-recorded material (for example AI-generated landing pages), we mark it as AI-generated. A person at the business reviews and approves every AI-drafted reply before it is sent, and the reply carries the disclosure line described above when the business sending it has turned it on. AI-generated landing pages also carry a machine-readable marker; machine-readable marking for other content produced through the product is in development to satisfy Article 50(2) of the AI Act.

Changes to this notice

We update this notice as the underlying AI features change. Changes that affect what data is processed or which providers are involved are surfaced separately to tenants in line with the subprocessor notice procedure above.

Contact

Questions about this notice, our AI features, or your AI Act rights: see the contact details in our Privacy Policy.