lightflow-rig-llm

# LightFlow RIG LLM Use `lightflow.rig_llm` to call an arbitrary LLM through LightFlow's RIG-backed runtime. ## Workflow - Workflow id: `lightflow.rig_llm` - Required inputs: `prompt` or `text`, and `model`. - Common inputs: `provider`, `system`, `preamble`, `temperature`, `max_tokens`, `additional_params`. - Outputs: `text`, `response`, `provider`, `model`. - Runtime capability: `lightflow.llm.generate`. - Runtime engine: `runner.v1`. - Node Schema: `provider` is a select control; `temperature` and `max_tokens` expose editor ranges; `additional_params` is a JSON editor. ## Runtime The workflow crate owns the Rig provider adapter and exposes it through its package runner. The LightFlow host does not need a Rig feature or provider implementation. For local contract checks, use `provider=mock` and any model name. Keep provider API keys only in runtime environment variables; secrets are not accepted as workflow inputs. ## Providers Set `provider` to one of: - `openai` or `openai-compatible` - `openai-responses` - `anthropic` - `ollama` - `openrouter` - `deepseek` - `xai` - `mock` for local tests Provider endpoints and secrets are administrator-owned environment configuration and are not accepted as workflow inputs. Provider secrets are read only from variables such as `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `OLLAMA_API_KEY`, `OPENROUTER_API_KEY`, `DEEPSEEK_API_KEY`, and `XAI_API_KEY`. OpenAI-compatible endpoints use the dedicated pair `OPENAI_COMPATIBLE_BASE_URL` and `OPENAI_COMPATIBLE_API_KEY`; they never receive `OPENAI_API_KEY`. Ollama may use `OLLAMA_API_BASE_URL`. `LIGHTFLOW_RIG_PROVIDER`, `LIGHTFLOW_RIG_MODEL`, and `LIGHTFLOW_RIG_SYSTEM` provide defaults. ## Examples Local mock verification: ```bash lfw run lightflow.rig_llm \ -i provider='"mock"' \ -i model='"fake-llm"' \ -i prompt='"hello"' ``` OpenAI-compatible endpoint: ```bash export OPENAI_COMPATIBLE_BASE_URL='https://gateway.example/v1' export OPENAI_COMPATIBLE_API_KEY='...' lfw run lightflow.rig_llm \ -i provider='"openai-compatible"' \ -i model='"gpt-4o-mini"' \ -i prompt='"Write a concise status update."' ``` Ollama: ```bash lfw run lightflow.rig_llm \ -i provider='"ollama"' \ -i model='"qwen2.5:14b"' \ -i prompt='"Summarize this workflow project."' ``` ## Validation ```bash lfw node test lightflow.rig_llm ``` ## API Usage Start `lfw serve`, then call the workflow through the shared HTTP run contract. Adjust `inputs` to match the workflow contract above. ```bash curl -sS -X POST http://127.0.0.1:5174/workflows/lightflow.rig_llm/run \ -H 'content-type: application/json' \ -d '{"inputs":{}}' ```