OpenAI LLM¶
Calls OpenAI chat/completions via Salt’s LiteLLM layer. You select an OpenAI model, provide prompts and generation settings, and the node returns the model’s text response. It includes built-in fallback mappings for common/legacy model names to help keep workflows working as models change over time.
Usage¶
Use this node whenever you need OpenAI-powered text generation in a workflow. Typical flow: prepare or construct your prompt (and optional system prompt), select an OpenAI model, tune temperature and max_tokens, then feed the output downstream (e.g., to parsers, tools, or display nodes). It is suited for general-purpose chat, writing, and reasoning tasks that require OpenAI models.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| model | True | STRING | The OpenAI model to use. A dropdown of available models is provided; legacy names are auto-mapped to current equivalents when possible. | gpt-4o |
| system_prompt | True | STRING | System-level instruction that sets behavior and role for the model. Keep concise and consistent across turns. | You are a concise technical assistant. |
| prompt | True | STRING | The main user prompt or message content sent to the model. | Summarize the following article in three bullet points. |
| temperature | True | FLOAT | Controls randomness. Lower is more deterministic; higher is more creative. | 0.7 |
| max_tokens | True | INT | Upper bound on the number of tokens generated in the response. Must be within the selected model’s limits. | 512 |
| input_1 | False | STRING | Optional additional input that you can interpolate into your prompt template or concatenate upstream. | Customer profile JSON |
| input_2 | False | STRING | Optional additional input for prompt composition. | Knowledge base excerpt |
| input_3 | False | STRING | Optional additional input for prompt composition. | Previous conversation context |
| input_4 | False | STRING | Optional additional input for prompt composition. | Formatting instructions |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| output | STRING | The OpenAI model's text response. | Here are three concise bullet points summarizing the article... |
Important Notes¶
- Model availability and naming can change; this node includes fallback mappings (e.g., GPT-4o mini → gpt-4o-mini) to reduce breakage, but you should periodically validate model selections.
- You must have a valid OpenAI API credential configured in your Salt environment or workspace settings for this node to run.
- max_tokens must respect the chosen model’s context window; if set too high, the service may error or truncate.
- Temperature interacts with prompts and model choice; for reproducibility, keep temperature low and prompts specific.
- Usage is subject to organization-level limits and quotas; heavy use may trigger rate limits.
Troubleshooting¶
- Model not found or deprecated: Pick a currently listed model from the dropdown or rely on the node’s automatic legacy-to-current mapping where available.
- Authentication error: Ensure an OpenAI API key is configured in your Salt credentials store or environment variables (e.g., set
in the appropriate location). - Rate limit or quota exceeded: Reduce request frequency, lower token usage, or upgrade plan/quotas; consider batching or adding delays.
- max_tokens too large: Decrease max_tokens to fit within the model’s context window and your prompt length.
- Empty or low-quality output: Lower temperature for more deterministic results, refine the system_prompt and prompt, and provide more concrete instructions or examples.