OpenAI LLM¶
Sends prompts to OpenAI language models through the Salt LiteLLM layer and returns the model’s response. Supports a wide range of current and legacy OpenAI model IDs with built-in fallback mapping. Designed to handle typical chat/completions use cases as well as newer OpenAI reasoning and small-footprint variants.
Usage¶
Use this node when you need text generation, chat-style responses, reasoning, or structured output from OpenAI models within a Salt workflow. Place it after any prompt-building or context-assembling nodes, then route its outputs to parsers, evaluators, or downstream application logic. Choose a model from the supported list (e.g., GPT-4o family, o-series, GPT-5 variants) and tune generation controls like temperature and max_tokens as needed.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| model | True | STRING | The OpenAI model identifier to use. Supports many current options and legacy names that are auto-mapped to current equivalents. | gpt-4o |
| input | True | STRING | Your prompt or message content sent to the model. Typically a user/system combined prompt or a prepared message string. | Summarize the following article in 3 bullet points. |
| temperature | False | NUMBER | Controls randomness. Higher values produce more diverse outputs. | 0.7 |
| top_p | False | NUMBER | Nucleus sampling parameter to control output diversity. | 0.9 |
| max_tokens | False | NUMBER | Maximum number of tokens to generate in the response. | 1024 |
| system_prompt | False | STRING | Optional system-level instructions to steer the assistant’s behavior. | You are a concise assistant that answers with bullet points. |
| response_format | False | STRING | Optional desired response format hint (e.g., plain text or JSON). | json |
| tools | False | Not specified | Optional tool/function definitions to enable tool calling if supported by the chosen model. | Not specified |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| text | STRING | The primary text output from the model. | • Point 1 • Point 2 • Point 3 |
| raw | JSON | Full raw response payload with metadata such as token usage and model info. | {"id":"resp_abc123","model":"gpt-4o","usage":{"total_tokens":345},"output":"..."} |
Important Notes¶
- • Provider is OpenAI. Display name appears as "OpenAI LLM" in the Salt UI.
- • Supported/fallback models include: gpt-4, gpt-3.5-turbo, gpt-4-turbo, gpt-4o, gpt-4o-mini, chatgpt-4o-latest, o1, o1-mini, o3, o3-mini, o4-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-5, gpt-5-chat-latest, gpt-5-nano.
- • Some legacy model names are automatically mapped to current equivalents in the UI.
- • This node is subject to LLM usage limits in Salt (per plan/remote config). If you hit limits, you may be blocked from adding/running more LLM nodes.
- • An OpenAI API key must be configured in your Salt environment by an administrator; the node itself does not take a secret key input.
Troubleshooting¶
- • Model not found or deprecated: Switch to a currently supported model (e.g., gpt-4o or gpt-4.1). The UI may auto-map some legacy names.
- • Rate limit or quota errors: Reduce request frequency, lower max_tokens, or wait for your quota window to reset. Verify your organization’s OpenAI billing and Salt plan limits.
- • Empty or low-quality output: Lower temperature for more deterministic results, provide clearer system instructions, or increase max_tokens.
- • Tool calling not working: Ensure the chosen model supports tools and that tool definitions are correctly structured. If issues persist, try another compatible model.
- • Node limit reached: Salt may restrict the number of LLM nodes per workflow/account tier. Remove unused LLM nodes or upgrade your plan.