Default Models
The Default Models tab sets the platform default models used by conversations. These defaults are the fallback the platform applies when a conversation does not name its own model for a given task, so picking sensible defaults here shapes file upload, retrieval, title generation, and memory across the whole organization. The tab is gated by the MAINTAIN_DEFAULT_MODELS permission and is the section Admin Settings opens on by default.
Snapshot of Protean AI PlatformOnly running model deployments can be selected. Make sure the deployment is running before you try to select it; deployments that are not running do not appear in the dropdowns. Each dropdown is filtered to the deployment kind that fits its role, and any selection can be cleared back to None.
The four selectors
There are four selectors, one per role. Each is a single dropdown populated only with running deployments of the matching kind, and each shows the chosen deployment's id as its value.
Embedding Model
The Embedding Model is the deployment used to generate and query vector documents. When a file is uploaded into a conversation, its contents are turned into embeddings with this model so they can later be searched. It matters because retrieval-augmented features depend on it: if no embedding model is set, files cannot be embedded, and file upload in conversations is disabled. Change it when you stand up a new embedding deployment or want all conversations to embed with a different model.
Ranking Model
The Ranking Model is the reranker used to reorder vector documents after an initial similarity search, so the most relevant chunks surface first. It is applied during retrieval on top of the embedding search to improve answer quality. It matters because, like the embedding model, it is required for file upload: without a ranking model the platform cannot rerank retrieved documents, so file upload in conversations is disabled. Set it whenever you want retrieval results reordered by a dedicated reranker.
Title Generation Model
The Title Generation Model is a chat-capable deployment used to generate a short title for a conversation from its contents. After a conversation starts, this model produces a readable title so conversations are easy to recognize in the sidebar. It matters because if no title generation model is set, conversations do not get auto-generated titles. Point it at a running chat deployment when you want automatic conversation titles.
Memory Model
The Memory Model is the default model used for memory extraction, the process that distills durable summaries and facts out of conversations. It is the model the platform applies when summary and fact settings do not specify their own model. It matters because memory quality depends on a capable model being available for extraction; set it to a running chat-capable deployment so memory features have a default to fall back on.
Saving
Save is enabled only once you make a change, and Cancel discards your edits and restores the last saved values. Saving writes the four defaults (embedding, ranking, title generation, and memory extraction) as the platform-wide defaults that conversations fall back to.
Keep the embedding and ranking selectors pointed at running deployments so file upload stays available, and use chat-capable deployments for the title generation and memory roles.