AI & Assistant Policy
LAST UPDATED: 27 MAY 202601. How we use AI in Triopia
Triopia uses AI models to support features such as briefings, summaries, drafts, translations, and transcriptions inside the workspace. AI is a tool that helps you prepare and review work faster; it is not a replacement for your judgment.
02. Model providers
We use selected AI model providers to power assistant features. Today this includes providers such as Anthropic and Google. We may add or change providers over time and will route specific tasks to whichever provider is best suited to the task.
03. No model training on your conversations
We do not train AI models on your conversations, messages, files, or other workspace content. Requests we send to AI model providers are sent for inference only, so they can generate a response for you in the moment.
04. What is sent to a model provider
When you trigger an assistant feature, the relevant context from your workspace is sent to a model provider over a secure connection so the provider can generate a response. We send only what is needed for that specific request.
05. You review before anything is sent externally
Assistant output is shown to you inside the workspace as a draft, briefing, or summary. Sending it as an email, SMS, fax, or other outbound communication requires your explicit action. The assistant does not send outbound communication on your behalf without your approval.
06. Accuracy and verification
AI models can make mistakes. Briefings can omit details. Drafts can misstate facts. Translations can lose nuance. Transcripts can mishear words. Review assistant output and verify critical details, including names, dates, amounts, and addresses, before relying on it.
07. Not professional advice
Assistant output is not legal, financial, medical, tax, or other professional advice. Consult a qualified professional before acting on information produced by the assistant.
08. Sensitive content
Do not paste extremely sensitive information, such as government identifiers or full payment card details, into the assistant unless it is necessary for the task you are completing. The less sensitive content you send to an AI model, the less risk in the unlikely event of a provider incident.
09. Limits on what the assistant will do
The assistant will not help with illegal activity, harassment, fraud, malware, or other harmful tasks. It may decline requests that would create safety, security, or legal risk.
10. Identity
The assistant identifies itself as an automated assistant when asked. It is not a person, and it will not claim to be one.
11. Tier-based availability
Some assistant features are available only on specific subscription tiers. See the pricing page for the capabilities included in each tier.
12. Improvements over time
We update the assistant as model providers ship new capabilities and as we learn from how customers use Triopia. We will not introduce changes that materially weaken safety or transparency without notice.
13. Reporting an issue
If the assistant produces output you believe is unsafe, incorrect, or inappropriate, report it through the support channel. We use reports to improve prompts, instructions, and safeguards inside Triopia.