Local LLM Automation: Generate Text Using On‑Prem Models

Local LLMs are becoming essential for privacy‑first automation — powering on‑premises agents, secure enterprise workflows, offline assistants, and edge‑AI deployments. The Local LLM Connector allows your workflows to generate text using a locally hosted model through an Ollama‑compatible HTTP API.

The connector definition states that it can "send a prompt to a local Ollama-compatible endpoint and return the generated text." It supports on‑prem execution only, making it ideal for environments where cloud AI is not allowed.

Step 1: Connect to a Local LLM Server

Drag a Generate Text node onto the canvas. This action sends a prompt to your local model server and returns the generated text.

Configure the following settings:

• endpoint — e.g., http://localhost:11434
• model — the model name (e.g., llama3, mistral, phi3)
• prompt — the main user prompt
• systemPrompt — optional system instruction
• temperature — creativity level (0.0–1.0)
• maxTokens — optional output length limit

The connector sends a JSON request to /api/generate and waits for a full response (streaming disabled).

Step 2: Generate Text

When the node runs, it constructs a request body containing:

• model — selected model
• prompt — user prompt
• system — optional system prompt
• options.temperature
• options.num_predict — max tokens

On success, the node returns:

• response — generated text
• model — model that produced the output
• ok — success flag

This is ideal for summarization, rewriting, classification, code generation, or any text‑based automation.

Step 3: Use Pipeline Input

The executor automatically merges upstream input when available. This allows you to:

• Generate text based on previous workflow steps
• Pass dynamic prompts from HTTP requests or triggers
• Build multi‑step reasoning chains

The input field is returned in the output for debugging and chaining.

Step 4: Handle Errors Gracefully

If the local LLM server is unreachable, misconfigured, or returns an invalid response, the node produces:

• ok = false
• error message — HTTP or JSON parsing error

This ensures workflows fail safely and predictably.

Step 5: Build On‑Prem AI Automations

Once generated text enters your workflow, you can combine it with any other connector:

• Send AI‑generated summaries to Email or Slack
• Store generated content in Google Sheets or databases
• Use LLM output to drive conditional logic
• Generate reports, descriptions, or structured data
• Build offline agents that run entirely on local hardware

Local LLMs become a powerful and private AI engine inside your automation ecosystem.

Example Workflow

Here’s a complete local‑AI automation you can build:

1. Receive a text input from an HTTP trigger.
2. Send it to a local LLM for summarization.
3. Store the summary in a database.
4. Send a Telegram notification with the generated text.
5. Append the result to a daily report file.

All of this runs offline — no cloud dependency, no external data exposure.

What's Next?

Try combining Local LLM with connectors like Filesystem, HTTP, Serial Bus, or OPC UA. Build intelligent agents, offline copilots, or private AI assistants that run entirely on your own hardware.

Local LLM unlocks powerful on‑prem AI — and this connector makes it simple, safe, and deeply integrated.