AI nodes: agents, models, sessions and channels

Ce contenu n’est pas encore disponible dans votre langue.

Generated on 2026-10-20 from a running server by tools/reference.mjs. Do not edit this page by hand.

The AI steps, and the resources you plug into them: models, sessions, channels to reach a person, policies and guardrails.

This page lists 23 nodes. How to read a node reference.

AI > Agents

Agent orchestration and execution.

Agent

Step · in the catalogue: AI > Agents · type agentic_node

Calls an agent of your library with a message, and gives its answer.

When to use it. To reuse an agent saved in your library: send it a message and get its answer back. Not for an agent built only for this workflow: use AI Agent instead.

Effects. costs, runs other things.

Fields

Name	Type	Required	Default	What it is
`entity_id`	Text	yes		The agent of your library to call.
`input`	Text	yes		Message sent to the agent.

Outputs

Name	Type	What it holds
`output`	Text	The final answer of the agent.

Any named output that the agent itself declares is added next to this one, under its own name.

Entries. In (Entry point)

Exits. Out (The answer of the agent), Error (Execution error if any)

AI Agent

Step · in the catalogue: AI > Agents · type ai.node

Runs a multi-step LLM agent with native tool calling, and searches the Knowledge Bases connected to it.

When to use it. To have an AI work through a task step by step in this workflow, with the model, tools, knowledge and people you plug into it. Not for a single question and answer: use LLM Prompt instead; to reuse an agent saved in your library, use Agent.

Effects. costs, runs other things.

Fields

Name	Type	Required	Default	What it is
`agent_mode`	Choice	no	`tools`	Agent execution mode. Choices: `tools`, `conversational`, `react`.
`system_prompt`	Text	no		System instructions (optional).
`prompt`	Text	yes	`input` (a reference)	User prompt (required).
`images`	ImageRef	no		Optional image to attach to the initial prompt — only used by vision-capable models (GPT-4o, Claude, Gemini). Ignored by text-only models.
`temperature`	Number	no	`0.2`	Creativity (0-2).
`max_iterations`	Number	no		Maximum number of rounds — one call to the model, then the tools it asked for (leave empty for no limit).
`response_quality_threshold`	Number	no	`0`	Minimum response quality score before finalization (0 disables the quality pass).
`tool_choice_mode`	Choice	no	`auto`	Tool usage strategy. Choices: `auto`, `required`, `none`.
`allow_parallel_tool_calls`	Boolean	no	`false`	Allow the model to ask for several tools in the same round.
`response_schema`	Text	no		JSON Schema for the final response (optional).
`allows_human_questions`	Boolean	no	`true`	Allow this agent to ask the user a question and wait for a reply. Turn off to guarantee it always delivers its final answer directly instead of pausing the run.
`session_id`	Text	no		Session identifier (leave empty to create a new one when a session resource is connected).
`session_scope_hint`	Text	no		Optional scope hint used by forked session resources.
`restore_session`	Boolean	no	`true`	Restore previous conversation history from the connected session resource.
`persist_session`	Boolean	no	`true`	Persist the conversation history to the connected session resource.

Outputs

Name	Type	What it holds
`response`	Text	Final agent response
`finish_reason`	Text	Final LLM finish reason
`iterations`	Number	Number of rounds the agent took
`tool_calls`	Text	Executed tool calls as JSON
`guardrail_warnings`	Text	JSON array of non-blocking guardrail matches (guardrail_id, stage, message) — branch on this with an If/Switch to decide what to do (empty array when nothing matched)
`session_id`	Text	Resolved session identifier when a session resource is connected
`session_scope`	Text	Resolved session scope
`session_restored`	Boolean	Whether an existing session history was restored
`session_message_count`	Number	Number of conversation messages currently stored in the session

Entries. In (Entry point)

Exits. Out (Default output port)

Places for resources

Place	Required	Accepts	How many
LLM Model	yes	`llm.model.*`	1
Knowledge Bases	no	`ai.knowledge.*`	several
Human Loop Channel	no	`interaction.channel.*`	1
Agent Session	no	`ai.node.session.*`	1
AI Agent Tools	no	`ai.node.tool.*`	several
AI Agent Policies	no	`ai.node.policy.*`	several
AI Agent Guardrails	no	`ai.node.guardrail.*`	several

Team

Step · in the catalogue: AI > Agents · type team_node

Calls a team of your library with a message, and gives its answer.

When to use it. To hand a task to a team of agents saved in your library and get its answer back. For a single agent, use Agent instead.

Effects. costs, runs other things.

Fields

Name	Type	Required	Default	What it is
`entity_id`	Text	yes		The team of your library to call.
`input`	Text	yes		Message sent to the team.

Outputs

Name	Type	What it holds
`output`	Text	The final answer of the team.

Any named output that the team itself declares is added next to this one, under its own name.

Entries. In (Entry point)

Exits. Out (The answer of the team), Error (Execution error if any)

AI > LLM

Text generation and prompting nodes.

LLM Prompt

Step · in the catalogue: AI > LLM · type ai.llm.prompt

Executes a prompt using an LLM with optional structured output, answering from the Knowledge Bases connected to it.

When to use it. To get a single answer from a model for one prompt (summarize, classify, extract, write), as text, or as JSON when you give a JSON Schema. Not for work that needs tools or several steps: use AI Agent instead.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`system_prompt`	Text	no		System instructions (optional).
`prompt`	Text	yes		User prompt (required).
`images`	ImageRef	no		Optional image to analyze — only used by vision-capable models (GPT-4o, Claude, Gemini). Ignored by text-only models.
`response_schema`	Text	no		JSON Schema for structured output (optional).
`temperature`	Number	no	`0.7`	Creativity (0-2).
`max_tokens`	Number	no	`2000`	Max response tokens.

Outputs

Name	Type	What it holds
`response`	Text	LLM response
`knowledge_used`	Text	What the Knowledge Bases gave for this prompt (empty when none is connected)

Places for resources

Place	Required	Accepts	How many
LLM Model	yes	`llm.model.*`	1
Knowledge Bases	no	`ai.knowledge.*`	several

AI > Models

LLM and image generation model resources.

Anthropic Model

Resource · in the catalogue: AI > Models · type ai.llm.model.anthropic

Anthropic Claude models (claude-opus-4, claude-sonnet-4, etc.).

When to use it. To give an AI Agent, an LLM Prompt or another AI step a Claude model, with a key from your Anthropic account. If your company reaches Claude through AWS, use AWS Bedrock Model instead.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Provider credentials. Connection of type `anthropic`.
`model`	Text	no		Model override (leave empty to use credential default).
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).
`frequency_penalty`	Number	no	`0`	Reduce repetition –2.0 to 2.0 (default: 0.0).
`presence_penalty`	Number	no	`0`	Encourage new topics –2.0 to 2.0 (default: 0.0).
`context_length`	Number	no		Context window override in tokens (LM Studio / Ollama only).

Plugs into a place that accepts llm.model.*.

AWS Bedrock Model

Resource · in the catalogue: AI > Models · type ai.llm.model.bedrock

Anthropic Claude models via AWS Bedrock — SigV4-signed requests through the official AWS SDK, no bearer token.

When to use it. To give an AI Agent, an LLM Prompt or another AI step a Claude model through your company’s AWS account, with AWS access keys. With a key from an Anthropic account, use Anthropic Model instead.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		AWS credentials used to sign Bedrock requests. Connection of type `aws_bedrock`.
`model`	Text	no	`anthropic.claude-3-5-sonnet-20241022-v2:0`	Bedrock model id.
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).

Plugs into a place that accepts llm.model.*.

DeepSeek Model

Resource · in the catalogue: AI > Models · type ai.llm.model.deepseek

DeepSeek AI models (deepseek-chat, deepseek-reasoner, etc.).

When to use it. To give an AI Agent, an LLM Prompt or another AI step a DeepSeek model, with a key from your DeepSeek account.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Provider credentials. Connection of type `deepseek`.
`model`	Text	no		Model override (leave empty to use credential default).
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).
`frequency_penalty`	Number	no	`0`	Reduce repetition –2.0 to 2.0 (default: 0.0).
`presence_penalty`	Number	no	`0`	Encourage new topics –2.0 to 2.0 (default: 0.0).
`context_length`	Number	no		Context window override in tokens (LM Studio / Ollama only).

Plugs into a place that accepts llm.model.*.

Google Gemini Model

Resource · in the catalogue: AI > Models · type ai.llm.model.gemini

Google Gemini models via AI Studio (gemini-2.0-flash, gemini-2.5-pro, etc.).

When to use it. To give an AI Agent, an LLM Prompt or another AI step a Google Gemini model, with an API key from Google AI Studio.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Provider credentials. Connection of type `gemini`.
`model`	Text	no		Model override (leave empty to use credential default).
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).
`frequency_penalty`	Number	no	`0`	Reduce repetition –2.0 to 2.0 (default: 0.0).
`presence_penalty`	Number	no	`0`	Encourage new topics –2.0 to 2.0 (default: 0.0).
`context_length`	Number	no		Context window override in tokens (LM Studio / Ollama only).

Plugs into a place that accepts llm.model.*.

Image Gen Model

Resource · in the catalogue: AI > Models · type imagegen.model

Exposes an image generation provider credential and model configuration.

When to use it. To choose the image provider, model, quality and style used by the AI Image Generator. Plug it into the AI Image Generator.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Image generation provider credentials (includes base_url). Connection of type `openai_image`.
`model`	Text	no	`gpt-image-1.5`	OpenAI image model to draw with (gpt-image family).
`quality`	Choice	no	`medium`	Image quality. Choices: `low`, `medium`, `high`.
`style`	Choice	no	`none`	Look of the image, said to the model with the description. Choices: `none`, `vivid`, `natural`.

Plugs into a place that accepts imagegen.model.*.

OpenAI Model

Resource · in the catalogue: AI > Models · type ai.llm.model.openai

OpenAI GPT models (and compatible: LM Studio, Ollama, Groq, Azure OpenAI…).

When to use it. To give an AI Agent, an LLM Prompt or another AI step a GPT model to work with. Also for services that follow the OpenAI API, such as Azure OpenAI, Groq, or a model on your own machine with LM Studio or Ollama.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Provider credentials. Connection of type `openai`.
`model`	Text	no		Model override (leave empty to use credential default).
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).
`frequency_penalty`	Number	no	`0`	Reduce repetition –2.0 to 2.0 (default: 0.0).
`presence_penalty`	Number	no	`0`	Encourage new topics –2.0 to 2.0 (default: 0.0).
`context_length`	Number	no		Context window override in tokens (LM Studio / Ollama only).

Plugs into a place that accepts llm.model.*.

xAI / Grok Model

Resource · in the catalogue: AI > Models · type ai.llm.model.xai

xAI Grok models (grok-3, grok-3-mini, etc.).

When to use it. To give an AI Agent, an LLM Prompt or another AI step a Grok model, with a key from your xAI account.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Provider credentials. Connection of type `xai`.
`model`	Text	no		Model override (leave empty to use credential default).
`max_tokens`	Number	no		Maximum tokens in response (leave empty to let the model decide).
`top_p`	Number	no	`1`	Nucleus sampling 0.0–1.0 (default: 1.0).
`frequency_penalty`	Number	no	`0`	Reduce repetition –2.0 to 2.0 (default: 0.0).
`presence_penalty`	Number	no	`0`	Encourage new topics –2.0 to 2.0 (default: 0.0).
`context_length`	Number	no		Context window override in tokens (LM Studio / Ollama only).

Plugs into a place that accepts llm.model.*.

AI > Sessions

Persistent agent session resources.

AI Agent Session (File)

Resource · in the catalogue: AI > Sessions · type ai.sessions.file

Stores AI agent conversation sessions as JSON files so they survive application restarts.

When to use it. To let an AI Agent remember the conversation from one run to the next, kept on disk so it survives a restart of the application. Plug it into the Agent Session of an AI Agent; give the same session ID to continue a conversation.

Effects. writes in the workspace.

Fields

Name	Type	Required	Default	What it is
`mode`	Choice	no	`isolated`	How the session is shared between connected agents. Choices: `isolated`, `shared`, `forked`.
`max_messages`	Number	no	`50`	Maximum number of conversation messages kept in the session.
`compaction_mode`	Choice	no	`none`	Optional compaction strategy applied before persistence. Choices: `none`, `summary`.
`tool_state_reuse_mode`	Choice	no	`none`	Optional reuse strategy for deterministic tool results stored in session state. Choices: `none`, `exact_match`.
`tool_state_reuse_ttl_seconds`	Number	no	`3600`	Maximum age in seconds for reusable tool-state entries.
`expiration_minutes`	Number	no	`0`	Optional session expiration in minutes (0 disables expiration).

Plugs into a place that accepts ai.node.session.*.

AI Agent Session (In-Memory)

Resource · in the catalogue: AI > Sessions · type ai.sessions.inmemory

Keeps AI agent conversations in memory until the application restarts.

When to use it. To let an AI Agent remember the conversation from one run to the next, until the application restarts. Plug it into the Agent Session of an AI Agent; to keep conversations after a restart, use AI Agent Session (File).

Fields

Name	Type	Required	Default	What it is
`mode`	Choice	no	`isolated`	How the session is shared between connected agents. Choices: `isolated`, `shared`, `forked`.
`max_messages`	Number	no	`50`	Maximum number of conversation messages kept in the session.
`compaction_mode`	Choice	no	`none`	Optional compaction strategy applied before persistence. Choices: `none`, `summary`.
`tool_state_reuse_mode`	Choice	no	`none`	Optional reuse strategy for deterministic tool results stored in session state. Choices: `none`, `exact_match`.
`tool_state_reuse_ttl_seconds`	Number	no	`3600`	Maximum age in seconds for reusable tool-state entries.
`expiration_minutes`	Number	no	`0`	Optional session expiration in minutes (0 disables expiration).

Plugs into a place that accepts ai.node.session.*.

AI > Interactions

Human-in-the-loop channels and interaction resources.

Chat Channel

Resource · in the catalogue: AI > Interactions · type ai.interactions.chat

Internal local chat channel for workflow conversations, human-in-the-loop responses, and AG-UI rich widgets.

When to use it. To let an AI Agent talk with the user in the app’s chat: send messages, ask a question and wait for the reply, show rich cards. Plug it into the Human Loop Channel of an AI Agent, or use it with an interaction step.

Effects. writes in the workspace.

Fields

Name	Type	Required	Default	What it is
`default_thread_title`	Text	no	`Main`	Default thread title for newly created workflow conversations.
`workflow_author_name`	Text	no	`Workflow`	Display name used for workflow-authored messages.
`respondent`	Text	no		Expected respondent name for traceability.
`timeout_minutes`	Number	no	`0`	Response timeout in minutes (0 = no timeout).
`widget_bundles`	TextList	no	`["agentivity_artifacts"]`	Platform widget bundle IDs to activate for this run.
`custom_widgets`	Structured	no	`[]`	Application-specific widget definitions the LLM can invoke as tool calls.

Plugs into a place that accepts interaction.channel.*.

Email Channel

Resource · in the catalogue: AI > Interactions · type ai.interactions.email

Email-based interaction channel for delivering requests and collecting responses.

When to use it. To tell someone by e-mail that a workflow needs their answer: the e-mail only notifies, the answer is still given in the app. An AI Agent cannot ask questions through it: use the Chat, Forms or Slack channel for that.

Effects. sends a message.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		SMTP credentials for sending emails. Connection of type `smtp`.
`to`	Text	yes		Recipient email address(es).
`reply_to`	Text	no		Reply-to address for responses.
`timeout_minutes`	Number	no	`1440`	Response timeout in minutes (0 = no timeout).

Plugs into a place that accepts interaction.channel.*.

Forms Channel

Resource · in the catalogue: AI > Interactions · type ai.interactions.forms

Web form-based interaction channel for collecting user input.

When to use it. To collect answers from a person through a form shown in the app, with only the fields needed. Plug it into an AI Agent so it can ask for missing information as a form and wait for it.

Effects. sends a message.

Fields

Name	Type	Required	Default	What it is
`respondent`	Text	no		Name or email of the person responding (for traceability in logs).
`timeout_minutes`	Number	no	`0`	Response timeout in minutes (0 = no timeout).

Plugs into a place that accepts interaction.channel.*.

Slack Channel

Resource · in the catalogue: AI > Interactions · type ai.interactions.slack

Slack workspace channel for delivering interaction requests and collecting responses.

When to use it. To ask people in Slack and take their reply from there, for an AI Agent or an interaction step. Needs a Slack bot token and the ID of the Slack channel, copied from Slack.

Effects. sends a message.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		Slack API credentials (Bot token). Connection of type `slack`.
`default_channel`	Text	yes		Default Slack channel ID for HIL interactions.
`mention_users`	Text	no		Users to mention (@user1, @user2).
`allowed_user_ids`	Text	no		Optional Slack user IDs allowed to answer (comma-separated).
`send_initial_message`	Boolean	no	`true`	Send the initial HIL prompt to Slack before waiting for a reply.
`timeout_minutes`	Number	no	`60`	Response timeout in minutes (0 = no timeout).

Plugs into a place that accepts interaction.channel.*.

AI > Policies

Composable AI agent policy resources.

AI Agent Policy

Resource · in the catalogue: AI > Policies · type ai.policies.layer

A layer of rules and preferences an AI agent follows, with its scope and its priority. Plug it into the Policies of an AI Agent.

When to use it. To give an AI Agent rules and preferences to follow, kept apart from its instructions so they can be reused. Plug it into the Policies of an AI Agent; when several layers apply, their scope and priority decide which comes first.

Fields

Name	Type	Required	Default	What it is
`scope`	Choice	no	`run`	Policy scope. Choices: `platform`, `resource`, `agent`, `mode`, `run`.
`priority`	Number	no	`1000`	Priority used during policy resolution.
`source_id`	Text	no		Optional policy source identifier.
`textual_policy_tree`	Structured	no		Structured textual policy tree for groups and policies.

Plugs into a place that accepts ai.node.policy.*.

AI > Guardrails

Composable AI agent guardrail resources.

AI Agent Guardrail

Resource · in the catalogue: AI > Guardrails · type ai.guardrails.keyword

Produces a simple keyword-based guardrail that can warn, rewrite, or block selected AI-agent runtime surfaces.

When to use it. To catch an exact word or phrase in what an AI Agent reads or writes, then warn, replace that text, or block. Plug it into the Guardrails of an AI Agent; to judge the meaning rather than exact words, use AI Semantic Guardrail.

Fields

Name	Type	Required	Default	What it is
`stage`	Choice	no	`final_response`	Guardrail stage. Choices: `planning_prompt`, `execution_prompt`, `task_batch_prompt`, `join_prompt`, `replanning_prompt`, `plan_quality_assessment_prompt`, `response_quality_assessment_prompt`, `response_quality_refinement_prompt`, `tool_failure_recovery_prompt`, `human_loop_duplicate_prompt`, `structured_response_finalization_prompt`, `conversational_finalization_prompt`, `agent_workspace_consolidation_prompt`, `human_loop_prompt`, `agent_workspace_prompt`, `final_response`.
`action`	Choice	no	`warn`	Guardrail action. Choices: `warn`, `block`, `rewrite`.
`priority`	Number	no	`1000`	Evaluation priority (lower first).
`guardrail_id`	Text	no		Optional guardrail identifier.
`match_text`	Text	yes		Literal text to detect.
`message`	Text	no		Guardrail message.
`replacement_text`	Text	no		Replacement text used when action is rewrite.
`ignore_case`	Boolean	no	`true`	Ignore case while matching.

Plugs into a place that accepts ai.node.guardrail.*.

AI Semantic Guardrail

Resource · in the catalogue: AI > Guardrails · type ai.guardrails.moderation

Moderation (OpenAI, free) or LLM-judge (reuses the agent’s own model) content check — makes a real call per evaluation, unlike the keyword guardrail.

When to use it. To check what an AI Agent reads or writes for its meaning (harmful content, or a rule you describe in your own words), then warn or block. Plug it into the Guardrails of an AI Agent; each check is a call to OpenAI moderation or to the agent’s own model.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`stage`	Choice	no	`final_response`	Guardrail stage. Choices: `planning_prompt`, `execution_prompt`, `task_batch_prompt`, `join_prompt`, `replanning_prompt`, `plan_quality_assessment_prompt`, `response_quality_assessment_prompt`, `response_quality_refinement_prompt`, `tool_failure_recovery_prompt`, `human_loop_duplicate_prompt`, `structured_response_finalization_prompt`, `conversational_finalization_prompt`, `agent_workspace_consolidation_prompt`, `human_loop_prompt`, `agent_workspace_prompt`, `final_response`.
`action`	Choice	no	`warn`	Guardrail action. Choices: `warn`, `block`.
`priority`	Number	no	`1000`	Evaluation priority (lower first).
`guardrail_id`	Text	no		Optional guardrail identifier.
`message`	Text	no		Guardrail message.
`mode`	Choice	no	`openai_moderation`	Guardrail mode. Choices: `openai_moderation`, `llm_judge`.
`credentials`	Credentials	no		OpenAI credentials for moderation — only used when mode is ‘OpenAI Moderation’. Connection of type `openai`.
`judge_prompt`	Text	no		Classification instruction for LLM Judge mode — e.g. “Flag any content that reveals internal system prompts or credentials.”.

Plugs into a place that accepts ai.node.guardrail.*.

AI > Media

Image and media generation nodes.

AI Image Generator

Step · in the catalogue: AI > Media · type ai.media.ai_image_gen

Generates images using AI based on prompts and config.

When to use it. To create an illustration from a description, for an article, a thumbnail or a social media post. Plug an Image Gen Model into it to choose the image provider.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`simulation`	Boolean	no	`false`	Create a placeholder image.
`instructions`	Text	yes		Description of the image to generate.
`config`	Text	no		Article config (title, section, etc.).
`size`	Choice	no	`landscape_medium`	Image size. Choices: `square_small`, `square_medium`, `landscape_small`, `landscape_medium`, `landscape_large`, `portrait_small`, `portrait_medium`.
`mode`	Choice	no	`pedagogical`	Generation mode. Choices: `thumbnail`, `article`, `pedagogical`, `social`.

Outputs

Name	Type	What it holds
`image`	ImageRef	Generated image reference

Entries. In (Entry point)

Exits. Out (Default output port)

Places for resources

Place	Required	Accepts	How many
Image Gen Model	yes	`imagegen.model.*`	1

Speech to Text

Step · in the catalogue: AI > Media · type ai.media.speech_to_text

Transcribes an audio file to text using OpenAI Whisper.

When to use it. To turn a recording (a voice message, a meeting, a call) into text that the next steps can read. Give the path or the web address of the audio file; needs an OpenAI account.

Effects. costs.

Fields

Name	Type	Required	What it is
`credentials`	Credentials	yes	OpenAI credentials (Whisper uses the same account as chat models). Connection of type `openai`.
`audio_source`	Text	yes	Local file path or http(s) URL to the audio file.
`language`	Text	no	ISO-639-1 language hint (leave empty for auto-detection).
`context_prompt`	Text	no	Optional context to improve accuracy (proper nouns, jargon, expected style).

Outputs

Name	Type	What it holds
`text`	Text	Transcribed text
`language`	Text	Detected language
`duration_seconds`	Number	Audio duration in seconds

Entries. In (Entry point)

Exits. Out (Default output port)

Text to Speech

Step · in the catalogue: AI > Media · type ai.media.text_to_speech

Synthesizes speech audio from text using the OpenAI TTS API.

When to use it. To turn a text into a spoken audio file, for example to read a summary aloud or produce a voice message. Not for transcribing a recording: use Speech to Text instead.

Effects. costs.

Fields

Name	Type	Required	Default	What it is
`credentials`	Credentials	yes		OpenAI credentials (TTS uses the same account as chat models). Connection of type `openai`.
`text`	Text	yes		Text to synthesize.
`voice`	Choice	no	`alloy`	Voice. Choices: `alloy`, `echo`, `fable`, `onyx`, `nova`, `shimmer`.
`model`	Choice	no	`tts-1`	Model. Choices: `tts-1`, `tts-1-hd`.
`response_format`	Choice	no	`mp3`	Audio format. Choices: `mp3`, `opus`, `aac`, `flac`, `wav`.

Outputs

Name	Type	What it holds
`audio`	AudioRef	Synthesized audio reference

Entries. In (Entry point)

Exits. Out (Default output port)