Flow Board

Start Node

The entry point of your flow. Automatically created when a flow is created. This is where execution begins.

Message Node

Displays a static message from the agent to the user. Can include variables like ${lastInput} or ${lastResponse}.

Prompt Node

Waits for user input. Stores the user's response in a flow variable (default: ${lastInput}). This is where conversations pause for user interaction.

Model Node

Calls an LLM (Large Language Model) with a system prompt. Processes user input and generates AI responses. Can be configured with temperature, model selection, and tools. Output is stored in a variable (default: ${lastResponse}).

Database Node

Performs a vector database search using embeddings. Searches a specified knowledge base with the provided search string. Returns relevant chunks that can be used in subsequent nodes.

Directory Category Listing Node

Displays a list of top-level categories from a selected knowledge base (with document counts). In the chat, the node title is shown as a header and categories appear in a compact multi-column layout; subcategories are shown one level deep with slight indentation. Categories with no documents are hidden.

Each category is a button. When the user clicks a category, a Directory Document Listing widget is added to the thread (no flow execution), showing a paginated list of documents in that category.

Directory Document Listing Node

When executed, outputs a widget that shows a paginated list of document cards from a selected knowledge base category. Configure the node with a knowledge base and a category; the chat displays the category name as a header and document cards (title, thumbnail, snippet, date) with previous/next pagination.

Intent Database Node

Special vector search that searches intent phrases across all flows in the agent. Used for intent detection and routing to appropriate flows.

Goto Flow Node

Navigates to another flow, allowing you to break complex conversations into reusable flow modules. Useful for creating modular conversation structures.

Condition Branch Node

Implements conditional logic to route to different branches based on input. Supports multiple matching methods:

• Regex Matching - Pattern-based text matching
• Cosine Similarity - Semantic similarity using embeddings
• LLM-based - Uses an LLM to classify input and select branch

Branch Node

Individual branch option for a Condition Branch Node. Each branch can have phrases associated with it for matching. Can optionally navigate to another flow.

Agent Memory Node

Manages user-specific memory stored in the agent's database. Supports three operations:

• GET - Retrieves a value from memory by key
• PUT - Stores a value in memory by key
• APPEND - Adds a child object to a key (for lists/arrays)

Avatar Node

Triggers actions for the avatar (toyHead). Allows you to control avatar behavior and animations during conversations.

Realtime Voice Node

Enables speech-to-speech conversations using the OpenAI Realtime API. Used for in-chat voice calls and for Twilio (SIP) phone calls when the agent is configured for voice. You set instructions, voice, and model in the inspector.

Tools: You can enable tools for the voice session so the assistant can run actions during the call (e.g. get date/time, create or search documents, manage tasks, show a URL, lead capture). Open "Configure Tools" in the inspector to turn built-in tools on or off. Knowledge base search and capture tools (update capture fields, update capture document, update capture fields with schema, update captured relationships) are added automatically when you connect the corresponding nodes to the Realtime Voice node with toolcall edges.

Tool execution runs on the server for both in-browser voice and phone calls; the client only starts the session and does not handle tool calls directly.