Coveo AI-Relevance Platform (Relevance Generative Answering, Passage Retrieval API, Agentic AI)

von Coveo

AI Application

Niedrige Vertrauensstufe

Coveo's SaaS AI-Relevance Platform unifies enterprise content and product catalogs in one secure index (with native connectors and item-level permissions) and applies per-customer machine learning (relevance tuning, query suggestions, content and product recommendations, semantic vector search, intent-aware product ranking) to power search, recommendations and personalization for commerce, customer service, websites and the digital workplace. Its generative layer, Relevance Generative Answering (RGA), retrieves the most relevant passages in two stages and sends a grounded prompt to a third-party LLM (OpenAI models via Microsoft Azure OpenAI, or Amazon Bedrock models such as Anthropic Claude, Meta Llama, AWS Nova and Titan) to return answers with citations. The Passage Retrieval API, Answer API and a hosted MCP server expose the same retrieval to customers' own RAG apps and agents (Salesforce Agentforce, SAP Joule, AWS AgentCore, ChatGPT Enterprise), alongside Conversational Search and Conversational Product Discovery. Rated medium: it processes ordinary enterprise content, end-user queries, prompts and behavioral clickstream rather than special-category data by design, permissions are enforced at retrieval, and the vendor has a strong assurance posture (SOC 2 Type II, ISO 27001, ISO 27018, ISO 27701, HIPAA option, public DPA and sub-processor list, documented 'zero retention' and no-training commitments for the shared LLM). The open points are data flows and use: Azure OpenAI is listed as a US-only sub-processor, so EU-hosted customers may see prompts and retrieved passages leave the EU; the DPA allows use of personal data for developing the Services (short of training models used by other customers); analytics data, including generative events, is kept 3 years; and customers can index sensitive content (emails, HR or health data under the HIPAA option). The rating would rise if the platform is used to answer questions in high-risk contexts (for example eligibility, employment or health decisions), if sensitive or special-category content is indexed and exposed to generation, or if agentic integrations are allowed to act without human review; it could fall to limited for public-website search without generative features.