AI Model¶

The AI Model sub-component provides LLM configuration to AI nodes. It holds the credential (API key) and model name that determine which language model an agent, categorizer, router, or extractor uses.

Ports¶

Inputs¶

This component has no inputs. It is a source-only node.

Outputs¶

Configuration¶

The AI Model node stores its configuration on the ComponentConfig record:

Both fields are required. An AI Model node without both values set will cause a build error when any connected AI node attempts to resolve its LLM.

Usage¶

1. Add an AI Model node from the Node Palette (Sub-components category)
2. Select an LLM credential from the dropdown in the node details panel
3. Select a model name from the available models (populated from the credential's provider)
4. Connect the AI Model to an AI node via the blue diamond model handle

Canvas layout¶

The AI Model node has a unique handle configuration:

• Top diamond handle (source) -- connects upward to AI nodes
• No bottom handles, no left/right circle handles

This means the AI Model always sits below or beside the AI node it serves, with the connection going upward.

One model per AI node¶

Each AI node requires exactly one AI Model connection. If multiple AI Model nodes are connected, the behavior depends on which edge the builder encounters first. Best practice is to connect exactly one AI Model per AI node.

Sharing models¶

A single AI Model node can be connected to multiple AI nodes on the same canvas. This is useful when you want several agents to use the same LLM:

flowchart TB
    A1[Agent A]
    A2[Agent B]
    A3[Categorizer]
    M[AI Model<br/>gpt-4] -.->|model| A1
    M -.->|model| A2
    M -.->|model| A3

LLM resolution¶

At build time, the resolve_llm_for_node() function in services/llm.py follows this path:

1. Find the AI Model node connected via an edge_label="llm" edge
2. Read the llm_credential and llm_model from its ComponentConfig
3. Load the credential from the database and decrypt the API key
4. Instantiate the appropriate LangChain LLM class (ChatOpenAI, ChatAnthropic, etc.)

Example¶

A basic chat workflow with an agent and its model:

flowchart LR
    T[Chat Trigger] --> A[Agent]
    M[AI Model<br/>claude-3-opus] -.->|model| A

The AI Model is configured with:

• Credential: anthropic-prod (an Anthropic API key from the Credentials page)
• Model: claude-3-opus-20240229

The Agent automatically uses this model for all LLM calls during execution.