AI Terms

The short version. Cord and the AI features use your data only to do what you asked. We do not train AI on your data, and every request we send asks for providers that neither train on it nor keep it. Speech is transcribed on our own servers; language and search models are reached through OpenRouter in the US. AI can be wrong, so check what it gives you. Admins can turn AI features off.

1. What uses AI

• Cord, the assistant: answers, summaries, drafts, actions you ask for, and memory you choose to keep.
• Meetings: live transcription, notes, outcomes and suggestions.
• Mail and sales: sorting replies and suggesting next steps.
• Search: finding things by meaning, not only by exact words.
• Voice: speech to text, for dictation, voice messages and meetings.

2. Who processes it

• On our servers. Transcription, voice and picture models run on our own servers at Hetzner in the EU. Audio does not leave them.
• Through OpenRouter. Language and search models are reached through OpenRouter, Inc. in the US, which passes each request to a provider that runs the model: Anthropic's Claude Haiku, run by Google Cloud (Vertex AI) or Amazon Web Services (Bedrock); Google's Gemini embeddings, run by Google; and TypeSafe AI's Jev decision model, for sorting. OpenRouter does not train on prompts or outputs.
• Our routing settings. Every request we send to OpenRouter tells it to use only providers that do not collect data for training, and only providers that do not retain prompts or outputs. If no provider meets both for a request, the request fails rather than going to one that does not.
• See the Subprocessors page for where each company is.

3. Your data, our promise

• No training. We do not use Customer Data, prompts or outputs to train or fine-tune any model, and our requests are routed only to providers that do not train on them.
• Only what is needed. Cord sends a model only the content a task needs, and only what the person asking may already see in the workspace. The Vault is never sent to AI.
• Yours. Prompts and outputs are Customer Data. As between us, you own the outputs, as far as the law allows them to be owned.

4. Using what AI gives you

AI output can be incomplete, wrong, or similar to output made for others. Review it before you rely on it, send it, or use it to make decisions about people. Do not use AI features for decisions with legal or similarly significant effects on people (for example hiring, firing or credit) without a person reviewing them. You must also follow the Acceptable Use Policy.

5. Being open about AI

Coding Cords marks AI-made notes, summaries and drafts in the product. Meeting participants are told when recording or transcription is on, and see it on screen while it is. Where something made by AI reaches people outside your company (for example a reply drafted by Cord and sent), you are responsible for telling them when the law requires it, including under the EU AI Act; we provide the labels and notices to help.

6. Control

Admins choose which AI features are on in Settings › AI, can turn transcription and recording off, and can see and delete what Cord remembers.