Foundation model
A large model pretrained on broad, general data that serves as a reusable base which downstream applications adapt through prompting, tool use, or fine-tuning.
In more detail
The term foundation model captures the idea that one large pretrained model can serve as the base for many different applications. Rather than train a new model per task, teams take a general-purpose foundation model and specialize it through prompting, retrieval, tool access, or light fine-tuning.
Foundation models are not limited to text. The same pretrain-then-adapt pattern applies to models for images, audio, and code. What they share is a broad base of general capability that downstream systems shape toward a specific job.
Where this shows up at Ceven
Ceven builds on foundation models rather than training its own. The value it adds sits above the model: turning a plain-language request into a running workflow, connecting it to the customer's tools, gating the risky steps for human approval, and recording everything. The foundation model supplies general language capability; Ceven supplies the orchestration that makes it do real work.