RealTimeX Setup
LLM Setup
Overview

LLM Setup

This section explains how to connect language model providers in RealTimeX.

Use it when you need to:

  • choose the default system LLM
  • connect a cloud or local provider
  • select the default model for that provider
  • find the provider-specific setup page for a backend you want to use

Open the main selector from Settings > AI Providers > LLM.

đź’ˇ

Use Large Language Models when you want the behavior guide for system defaults, workspace overrides, and provider strategy. This section is for connection and setup.

What this setting controls

The LLM page is the instance-wide source of truth for the default chat model provider.

From that page, you configure:

  • the system default provider
  • provider credentials, endpoints, or base URLs
  • the default model used by that provider
  • plugin-provided LLM providers when plugins register them
  • the provider list used by workspace and agent override pickers

Common setup flow

Most providers follow the same setup pattern:

  1. Open Settings > AI Providers > LLM.
  2. Select the provider you want to use.
  3. Enter the required connection details.
  4. Choose a model, or type a model ID when the provider expects manual entry.
  5. Save the configuration.
  6. Test the result in a normal chat before using it in workspaces, agents, or automations.

Depending on the provider, RealTimeX may load the model list only after you enter a valid API key, endpoint, or base URL.

How provider selection layers work

System default

The system LLM is the fallback provider for the instance.

If nothing more specific is selected, RealTimeX uses this provider for:

  • regular chats
  • new workspaces that inherit the default
  • tools or automations that do not define their own provider

Workspace override

Workspaces can stay on System default or choose a different provider and model for that workspace only.

Use workspace overrides when:

  • one team needs a different vendor
  • one workspace needs a larger context window
  • one project should stay on local inference while the rest of the instance uses cloud models

Agent and runtime override

Some agent surfaces can override the inherited provider and model as well.

If an agent or runtime does not define its own LLM, it falls back to the workspace or system selection, depending on that feature's configuration.

Which provider type should you choose?

RealTimeX-managed options

  • RealTimeX Cloud is the simplest hosted path when you want a managed cloud provider inside RealTimeX.
  • RealTimeX Local is the integrated local GGUF workflow when you want on-device inference without running a separate model server yourself.

RealTimeX Cloud is configured directly from the same LLM selector and normally does not require a separate setup flow beyond choosing the provider and model.

External cloud providers

Use providers like OpenAI, Anthropic, Gemini, Azure OpenAI, Groq, or Mistral when your team already operates around that vendor or you want access to that model family directly.

Local or self-hosted endpoints

Use providers like Ollama, LM Studio, Local AI, KoboldCPP, llama.cpp, or Generic OpenAI when you already run local inference infrastructure or a self-hosted gateway.

Compatibility and proxy layers

Use Generic OpenAI or LiteLLM when you want RealTimeX to connect through an OpenAI-compatible proxy or compatibility layer instead of a vendor-specific integration.

Plugin providers

Plugins can register additional LLM providers. When that happens, they appear in the same selector as built-in providers and follow the same general setup flow.

Local and managed setup guides

Cloud setup guides

OpenAI
Connect a standard OpenAI API key and choose an OpenAI chat model.
Azure OpenAI
Use Azure-hosted OpenAI deployments with endpoint and deployment configuration.
Anthropic
Connect Claude models directly through Anthropic.
Google Gemini
Use Gemini models through Google's hosted API.
Groq
Connect Groq-hosted models for low-latency inference.
Mistral AI
Use Mistral-hosted models through the Mistral API.
Cohere
Connect Cohere command models through a hosted API key.
OpenRouter
Route model access through OpenRouter's multi-provider catalog.
Together AI
Use Together-hosted open model APIs inside RealTimeX.
Perplexity AI
Connect hosted Perplexity models where that provider fits your workflow.
AWS Bedrock
Use foundation models through your AWS Bedrock account.
Hugging Face
Point RealTimeX at a Hugging Face Inference Endpoint.

Compatibility and gateway guides

Providers you may also see in the app

The current in-app selector can include additional built-in providers and runtimes that do not all have dedicated setup guides in this section yet, including:

  • DeepSeek
  • Fireworks AI
  • Novita AI
  • Moonshot AI
  • xAI
  • PPIO
  • NVIDIA NIM
  • LiteLLM
  • APIpie
  • llama.cpp
  • Oobabooga Web UI
  • Dell Pro AI Studio

Those still follow the same core pattern:

  • choose the provider
  • enter the required API key or endpoint
  • select or enter a model
  • save and test

Related guides