Automata: Internet of Agents

Tips & Tricks
1

Hi everyone,

I’ve published a new set of community nodes that integrate n8n with an intent-based agent registry called Automata.

The idea is simple:

Instead of calling fixed HTTP endpoints, your workflows can discover and call agents by intent, with ranking, execution keys and feedback – using dedicated nodes.

npm package: @apptrix/n8n-nodes-automata
GitHub repo (project + SDKs + registry): GitHub - MendesCorporation/automata: Automata is an “internet of agents” toolkit: a registry that ranks/authenticates agents, plus Provider and Consumer SDKs. Providers auto-register HTTP agents with intents/tasks and validate execution keys locally; Consumers use LLMs to analyze prompts, search ranked agents, build params from schemas, execute, and send feedback.

What is Automata?

Automata is an open-source “internet of agents” experiment:

  • Registry Central (Fastify + PostgreSQL) that stores agents and ranks them
  • Provider SDK to register agents with:
    • intents, categories, tags, locationScope, input_schema, etc.
  • Consumer SDK to:
    • analyze prompts with an LLM
    • search agents by intent
    • execute them with short-lived JWT execution keys
    • send back feedback (success / latency / rating)

The n8n nodes sit on top of this and let you use the registry without writing custom code.

Included nodes

  1. Automata Provider Webhook

    • Fixed path: automata-provider/execute
    • Registers a webhook in Automata when the node is activated
    • Validates the incoming JWT using your provider secret
    • Exposes the request payload to the workflow
    • Optional: delayed response using the “Respond” node (useful for long-running flows)

  2. Automata Respond

    • Sends the HTTP response back to Automata
    • By default wraps the data into { success: true, data }
    • Can also send error payloads when needed

  3. Automata Consumer

    • Authenticates with the registry (Dev / Prod / Custom URL)
    • Optionally uses an LLM (OpenAI by default) to:
      • analyze a natural language prompt
      • search for agents by intent / categories / tags / location
    • Executes one or multiple agents with execution keys
    • Sends feedback automatically (success + latency + optional rating)
    • Returns:
      • raw agent responses
      • intermediate analysis (intents, categories, location, etc.)

  4. Automata Consumer Tool

    • Same parameters as the Consumer node
    • Exposed as a tool so you can plug it into an ai_agent → ai_tool setup in n8n

Example flows

1) Turn an n8n workflow into an “Agent”

  • Drag Automata Provider Webhook
  • Build your workflow (call APIs, databases, etc.)
  • End with Automata Respond

Now this workflow:

  • is registered as an agent in the registry
  • has one or more intents (e.g. travel.hotel.search)
  • can be discovered and called by any Automata Consumer (SDK, n8n, or other services)

2) Agent-Orchestrated Quotes

You can create a flow that:

  1. Receives a natural language request (HTTP trigger, form, etc.)
  2. Passes it to Automata Consumer:
    • “Find development agencies that can quote this project”
  3. The node:
    • analyzes the prompt (LLM)
    • searches agents with intent development.agency.quote
    • executes multiple agents in parallel with executeMultipleWithFeedback
  4. You then:
    • sort and filter results in n8n
    • send the best quote via email / Slack / WhatsApp, etc.

Installation

In n8n:

  • Go to Settings → Community Nodes
  • Click Install
  • Enter: @apptrix/n8n-nodes-automata

You will also need:

  • an Automata Registry URL (you can self-host from the GitHub repo or use the dev instance)
  • a provider secret / consumer credentials (documented in the repo)

Looking for feedback

This is still early stage, but already usable. I’d love feedback on:

  • node UX (parameters, naming, defaults)
  • better ways to integrate with ai_agent / ai_tool
  • ideas for presets (e.g., common registries, starter workflows)
  • any security concerns or improvements you’d suggest

If anyone is interested, I can also share full example workflows (JSON) here.

Thanks!