Palatine Speech icon

Palatine Speech

Transcribe audio, diarize speakers, analyze sentiment, generate AI summaries, or use OpenAI-compatible LLM Chat Completions via Palatine. Keywords: n8n-community-node-package, n8n, palatine, speech-to-text, transcribe, transcription, stt, audio, ai, automation, voice-to-text, speech-recognition, audio-transcription, audio2text, audio-processing, diarization, speaker-diarization, speaker-segmentation, summarization, audio-summarization, sentiment-analysis, emotion-detection, tone-analysis, llm, chat-completions, openai-compatible, structured-output, json-schema

Actions6

LLM → Message a Model

AI-generated

Summary

Send a user message and optional system prompt to a Palatine LLM to generate a text or structured JSON response. Choose a model, enable thinking if needed, and optionally control sampling and completion length.

Inputs

  • llmModel (required) — Required model selected from the live model list or entered as an ID, such as palatine:m. Empty or whitespace-only values are rejected.
  • thinking — Enable extended reasoning mode; defaults to false.
  • responseFormatType — Choose text (default), JSON Object, or JSON Schema. This controls both the requested format and the output fields.
  • responseJsonSchema — Used for JSON Schema responses. Must be valid JSON containing a string name and a schema object; invalid configuration stops the request.
  • systemPrompt — Optional instructions for model behavior and tone. Trimmed and sent as a system message only when nonempty.
  • userMessage (required) — Required message to send to the model. Trimmed before sending; empty or whitespace-only messages are rejected.
  • llmOptions — Optional Max Tokens (1–32768) and Temperature (0–2). Sent only when supplied; the UI offers defaults of 512 and 1 respectively.
  • requestOptions — Declared request settings for batching, accepting invalid SSL certificates, an HTTP proxy, and timeout in milliseconds.

Output shape

JSON items whose fields depend on Response Format Type, with incoming binary data preserved.

Text returns {response}, using the first choice's message content. JSON Object returns id, model, created, choices, and usage; content remains inside choices and is not parsed separately. JSON Schema returns response, parsed, schema_valid, and schema_errors. Parsing accepts surrounding code fences; parsing or validation failures are reported in these fields rather than thrown. Schema validation is limited to basic object, array, and primitive types, required properties, and additionalProperties:false, with a recursion depth limit.

Links

Discussion