Model Integration & Configuration

GeniSpace connects to a range of AI models so your agents can run on the model that best fits your needs. The set of available models is managed by your platform administrator and can differ between the SaaS editions and self-hosted deployments.

How Models Work in GeniSpace

• One model per agent: Each agent is bound to a single model. You choose the model when creating or editing the agent, and it is used for that agent's conversations and task execution. (There is no automatic, per-request switching between models.)
• Models are administrator-configured: The model catalog is stored in the platform database. The list you see in the agent editor comes from the platform's models registry (GET /models), so the exact models available depend on what your administrator has configured for your deployment.
• Platform default: When a supported creation or built-in-agent flow does not specify a model, the platform resolves the administrator-configured default model. Do not assume a fixed identifier across editions or deployments.

info

The model lists on this page are examples only. The actual models available in your environment are configured per deployment by your administrator.

Supported Model Providers

GeniSpace integrates models through provider-specific adapters. The following provider types (API types) are supported:

Provider Type	Notes
OpenAI-compatible	Models exposed through an OpenAI-compatible contract
Alibaba Cloud / DashScope	Qwen and supported Alibaba Cloud models
Anthropic-compatible	Claude models when configured
Google-compatible	Gemini models when configured
Self-hosted gateway	Models explicitly registered by the administrator

The actual provider/channel types and capabilities are defined by the Models service catalog. An unknown provider is not automatically guaranteed to work.

Example Models

The following are examples of models that may be configured in a deployment. Availability is admin-configured per deployment and is not guaranteed.

• Qwen (Tongyi Qianwen) series — for example qwen-max, qwen-plus, qwen-turbo
• OpenAI GPT models
• Anthropic Claude models
• DeepSeek models
• Google Gemini models (via Vertex AI)
• Self-hosted / open-source models served through Ollama or Hugging Face endpoints

Deployment Options

SaaS Cloud Service

The GeniSpace SaaS editions provide ready-to-use cloud services:

• No maintenance: Model availability and upgrades are handled by the GeniSpace team
• Ready to use: Select an available model when configuring an agent
• Regional editions: The China and International editions may offer different model sets

Self-Hosted Deployment

GeniSpace self-hosted (Standalone) deployments let you connect models within your own environment:

• Data control: Run agents against models hosted in your own infrastructure
• Custom integration: Connect to your existing AI infrastructure (for example, via the ollama or OpenAI-compatible provider types)
• Bring your own credentials: Configure provider credentials as part of your deployment configuration (see below)

Model Selection for Agents

When you create or edit an agent, you select one model from the list of models available in your deployment. Different models have different strengths, so you may pick different models for different agents:

Use Case	Example Model Characteristics
Creative content generation	Models with strong generative/creative ability
Code generation	Models optimized for code and reasoning
Data analysis	Models with strong structured-output and reasoning ability
Simple Q&A	Lighter-weight, lower-cost models
Multilingual processing	Models with strong cross-language ability

The specific models that fit each use case depend on what is available in your deployment.

Model Connection & Credentials

Model credentials and routing are managed by Models service channels and the platform model catalog. End users select an available model; they do not receive provider credentials.

In short:

1. Administrators register a routable model and its capabilities.
2. Models service stores provider routing and credentials in protected administration data.
3. The platform maps user or Space context into the Models request for authorization, usage, and billing.
4. End users select from the enabled catalog.

RAG Knowledge Base Integration

GeniSpace provides a Retrieval-Augmented Generation (RAG) capability so agents can ground their answers in your own knowledge:

Knowledge Base Construction

1. Multi-source data ingestion: Supports documents and other data sources
2. Document processing: Processes document structure and extracts content
3. Vectorization: Uses embedding models to generate semantic vectors
4. Incremental updates: Supports updating the knowledge base over time

Retrieval-Augmented Generation

1. Semantic retrieval: Retrieves relevant content for the agent
2. Context injection: Adds retrieved content to the prompt context
3. Source grounding: Helps attribute generated content to its knowledge sources

Knowledge-base embedding and retrieval run through the platform model gateway. Dataset vectorization is a separate Dataset capability and uses the model bound to that Dataset. See Data & Knowledge Base for details.

Cost Management

All AI services consume quota from your account balance. To manage costs:

1. Monitor usage — Track quota consumption in the Analytics module
2. Choose appropriate models — Use lighter-weight models for simple tasks and reserve more capable models for demanding work
3. Optimize prompts and parameters — Concise prompts and reasonable token limits reduce per-call consumption

See Billing & Subscription for how quota and billing work.

FAQ

Can a single agent use more than one model?

No. Each agent is bound to a single model. If you want different behavior, create separate agents with different models, or change the model on an existing agent in its configuration.

Which models are available to me?

The models you can choose come from your deployment's model registry, which is configured by your administrator. The lists in this document are examples and may not match your environment. Check the model selector in the agent editor for the authoritative list.

How are provider credentials kept secure?

Provider credentials are stored in protected Models service channel configuration and are not exposed to end users, who only select from the enabled catalog.

Next Steps

• Learn about model selection in Agent Overview
• Study Data & Knowledge Base
• Master Prompt Engineering Best Practices