Cortex
Cortex features
Every capability that ships with Cortex — grouped by the job it does. Two agent types, citation-backed grounding, enterprise governance, model choice, and delivery paths from internal tool to embedded product.
Agents
Two agent types, one platform
Answer strictly from the sources you connect, with a citation on every response.
Codify multi-step expert workflows so anyone on the team can run them consistently.
Chain agents together — a Knowledge Agent can feed a Skills Agent that produces a deliverable.
Grounding
Grounded in your knowledge
Every fact links to the exact document and passage it came from.
PDF, Word, PowerPoint, Excel, SharePoint, Notion, Drive, S3/GCS, databases, transcripts, audio and video.
Schedule re-indexing per source; agents always answer from the current knowledge state.
Governance
Enterprise-grade by default
Shipped on day one — not gated behind an enterprise-tier upsell.
GDPR compliant. SOC 2 in progress. Your IP never trains a public model.
Scope who can query, edit, or deploy each agent — down to a single user.
Model choice
Model-agnostic by design
Route each agent to the foundation model that fits — GPT, Claude, Llama, Mistral, and more.
Change the underlying model and keep the agent, sources and evaluations intact.
Run open-weight models fully inside your Swiss/EU boundary when required.
Delivery
From workspace to product
Domain experts ship agents themselves — no engineering backlog required.
Ship agents in your own product, under your own brand, or expose them via API.
Preview, publish, iterate — deploy behaviour and content changes without redeploy cycles.