Available on: All plans (system defaults) / Starter+ (custom models) | Role: Organization Owner | Status: Generally Available
Overview
The AI Models section lets the Organization Owner register, validate, and manage AI models from multiple providers. Once registered, models become available for assignment to teamspaces and AI Boxes across the organization.System Default vs. Custom Models
Every organization starts with system default models provided by the platform. Organization Owners on Starter+ plans can add custom models from supported providers.| System Default | Custom | |
|---|---|---|
| Available on Free plan | Yes | No |
| Managed by | Platform | Organization Owner |
| Deletable | No | Yes |
| Configurable | No | Yes |
Free plan organizations have access to system default models only. Custom model registration requires a Starter plan or higher.
Supported Providers
Singularity AI supports models from the following providers:OpenAI
GPT-5, GPT-4.1, GPT-4o
Anthropic
Claude Opus, Claude Sonnet
Azure OpenAI
Azure-deployed OpenAI models
Google AI Studio
Gemini models
AWS Bedrock
AWS-hosted models
Mistral
Mistral models
Model Capabilities
Models are tagged with capabilities that determine where they can be used:| Capability | Description |
|---|---|
| Conversation | General chat and Q&A |
| Streaming | Real-time response streaming |
| Vision | Image understanding and analysis |
| Embeddings | Document vectorization for RAG (separate from chat models) |
| Function Calling | Tool use and API integration |
| Moderation | Content filtering |
How Models Are Used
Once registered, models become available across the platform:- Teamspace assignment — Admins assign models to teamspaces for chat conversations. Each teamspace can have multiple models (one per capability: chat, embeddings, vision, etc.)
- Connection embedding — Each data connection uses an embedding model to vectorize documents for AI search
- AI Box execution — Models can be selected per-box for LLM-powered workflow steps
- Knowledge Brain processing — Brains use models for knowledge extraction, scoring, and validation