AI transparency
How Syncanix discloses AI use to end users in every language, which models power the assistant, and how answers are grounded in your content.
Syncanix is open about being AI. End users always know they’re talking to an assistant, and you always know which models are doing the work.
Disclosing AI to end users
The chat widget tells end users they’re talking to AI, in keeping with transparency obligations such as the EU AI Act’s Article 50. The disclosure appears in the user’s own language across all six supported languages.
Which models we use
Syncanix publishes model cards describing each model in the system:
- Chat — Anthropic Claude (with OpenAI failover)
- The conversation is powered by Anthropic’s Claude under managed access, with an OpenAI model as failover for resilience.
- Retrieval — Amazon Bedrock
- Search over your knowledge uses Amazon Bedrock for embeddings and reranking, in-region.
- Your own models (BYOK)
- On eligible plans you can route the language model through your own provider key, so inference runs against your account.
Grounding answers in your content
When you add knowledge, the assistant grounds its answers in your material rather than guessing, and retrieval runs in-region. The result is answers that reflect your product, with the sources behind them.