Anthropic LLM¶
Sends a text prompt to Anthropic models via the LiteLLM integration and returns the model’s generated text. You select an Anthropic model, provide a system prompt and user prompt, and control generation with temperature and max token settings. Includes fallback model mappings to keep workflows running if a preferred model name isn’t available.
Usage¶
Use this node whenever you need text generation from Anthropic’s Claude family within a Salt workflow. Typical usage: set the model (e.g., a Claude Sonnet/Haiku/Opus variant), provide a system prompt to steer behavior, pass the main prompt (optionally augmented by auxiliary inputs), and adjust temperature/max tokens. Chain the output into downstream nodes for further processing or display.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| model | True | STRING | The Anthropic model to use. Choose from available options; the node provides fallback mappings for common Claude variants if direct names are not available. | claude-3-5-sonnet-20241022 |
| system_prompt | True | STRING | High-level instructions defining the assistant’s role, style, or constraints. Applied before the main prompt to steer the model’s behavior. | You are a concise assistant that answers with clear bullet points. |
| prompt | True | STRING | The main user input or task to complete. This is the content the model will respond to. | Summarize the following article in 5 bullets: |
| temperature | True | FLOAT | Controls randomness of the output. Lower values make responses more deterministic; higher values make them more creative. | 0.5 |
| max_tokens | True | INT | Maximum number of tokens to generate in the response. Actual limits may depend on the selected model. | 1024 |
| input_1 | False | STRING | Optional auxiliary input to provide extra context or variables to your prompt. | Customer profile data JSON |
| input_2 | False | STRING | Optional auxiliary input to provide extra context or variables to your prompt. | Conversation history text |
| input_3 | False | STRING | Optional auxiliary input to provide extra context or variables to your prompt. | Knowledge base excerpt |
| input_4 | False | STRING | Optional auxiliary input to provide extra context or variables to your prompt. | Task parameters JSON |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| Output | STRING | The text generated by the selected Anthropic model. | Here are the five key points from the article... |
Important Notes¶
- Model selection: The node exposes Anthropic models and includes fallback mappings (e.g., various Claude Haiku/Sonnet/Opus versions). If an exact name isn’t available at runtime, a mapped alternative may be used.
- Token limits: Effective max token limits depend on the chosen model. If you set max_tokens above a model’s allowance, the service may cap it or return an error.
- Temperature scale: Temperature is on a 0–1 scale; lower values yield more deterministic outputs.
- Auxiliary inputs: input_1 to input_4 are optional and can be used to pass additional context for your prompts.
- Service configuration: Access to Anthropic models depends on your Salt environment’s configured providers and credentials. Ensure Anthropic access is enabled by your admin.
Troubleshooting¶
- Model not found: If a selected model isn’t available, choose another from the list or rely on the provided fallback mappings.
- Empty or truncated output: Reduce temperature for stability, lower max_tokens if hitting limits, or simplify/shorten your prompt.
- Provider/credential errors: Verify that Anthropic access is configured in your environment and that your organization has the necessary permissions.
- Inconsistent style or behavior: Strengthen the system_prompt with clearer, explicit instructions, or reduce temperature.
- Long-running or timeout: Decrease max_tokens, simplify prompts, or try a smaller/faster model variant.