Default Model Configuration

Set org defaults for 6 capability types (LLM / Embedding / Reranking / Image / Audio / Video). When builders don't specify a model, they inherit these defaults.

Defaults for 6 Capabilities

Capability	Used by
LLM	Agent · Chatflow · Workflow LLM nodes
Embedding	Knowledge base default vectorization model
Reranking	Default reranker for knowledge retrieval nodes
Image	Image generation scenarios
Audio	Speech synthesis / recognition
Video	Video generation

Three-Layer Inheritance & Override

Org default (Layer 3, this page)
   ↓ overridable by
Workspace default (Layer 2, workspace settings)
   ↓ overridable by
Agent / Workflow node explicit (Layer 2, edit)

Nearest wins: Agent explicit > Workspace default > Org default.

Three Typical Patterns

1 · Simple SaaS

Org defaults all use GPT-4 / Claude; no workspace overrides → uniform standard org-wide.

2 · Private

Org default is self-hosted Qwen / Llama → fully on-prem org-wide.

3 · Hybrid (cost-sensitive / quality-sensitive)

Org default: Claude Sonnet (medium)
Workspace A default: GPT-4 (quality first, customer service)
Workspace B default: Llama 3 70B self-hosted (cost first, internal research)

Switching Defaults Doesn't Apply Retroactively

Already published Agents / Workflows do not auto-switch

Default changes affect only newly created apps and nodes that don't explicitly specify a model. Already explicit nodes stay the same.

For bulk switching, edit them in the workspace one by one or use the API.

Next Steps

Register a model → Model definition
Routing → Interface platform

Default Model Configuration

On this page