Streaming Hub
Streaming Hub 0.1

AI Services

Configure AI providers and use their models in Streaming Hub automations.

Alpha
Verified withStreaming Hub: 0.1.3.0Plugins: None

Connections and model profiles

Open Project Settings → AI Providers. A provider connection stores the endpoint and credentials. A model profile then chooses a connection, model and request defaults for AI steps.

Supported drivers

DriverUse
OpenAI CompatibleServices that expose compatible chat-completions and model-list endpoints, including a configured LM Studio server.
OllamaOllama chat and model-list endpoints.
Anthropic ClaudeAnthropic Messages API and model listing.

Add a provider connection

  1. On Provider settings, select Add item and give the connection a name.
  2. Choose a preset, or select a driver and enter the provider's Base URL and API key where required. Keep Enabled selected.
  3. Select Apply, then Ping Connections to check connectivity and Fetch Models to load the model list.
Provider settings keeps endpoint and credential details in connections. Use Ping Connections and Fetch Models before choosing a model profile.

Choose a model profile

  1. Open Model settings and select Add item.
  2. Choose the connection and a model returned by that provider.
  3. Adjust Response Timeout or Context Window Tokens only when the chosen provider or model needs them, then select Save.
A model profile selects a connection and model; its optional controls apply to requests made through that profile.

LM Studio

Run the LM Studio local server and choose the LM Studio preset for an OpenAI Compatible connection. The screenshot uses http://127.0.0.1:1234/v1; the address must match your local server configuration.

Automation use

AI Steps can send messages, select response format behavior and store returned text in Runtime variables or project variables. Parameters such as temperature, maximum tokens and top-p are translated according to the selected driver.

Context and privacy

Review prompts, context files and conversation history before sending a request. A remote provider receives the content included in that request. See Execution context for the difference between per-run data and persistent storage.

Limitations

Compatible providers can implement only part of an API. Model listing, structured response formats or thinking controls may not be supported by every server or model.

Troubleshooting

Verify that the provider server is running, the base URL is reachable, the selected model exists and required authentication headers are present. Inspect AI entries in the Log panel for HTTP status and timeout information.

Related