n8n-nodes-ollaya

Ask typed questions about any text or JSON from an n8n workflow, and route items by the answer, using decision models that return calibrated probabilities.

This community node connects n8n to two interchangeable backends that speak the same API:

The node is compatible with TypeSafe AI: it uses the same /v1/systemone and /v1/models endpoints, so a workflow built against the TypeSafe cloud runs against a local Ollaya server by changing one field in the credential, and back.

This is an independent community node, published by Cogniwork. It started as a fork of @typesafe-ai/n8n-nodes-typesafe-ai (MIT) and is not affiliated with or endorsed by TypeSafe AI or by the Ollaya project.

What is a decision model?

A decision model reads a state (a support ticket, an email, a chat message, any JSON) together with typed questions, and answers all of them in a single forward pass, in milliseconds. It never writes free text: each answer is a choice, a score or a probability, with a confidence you can act on. That makes it a good fit for the parts of a workflow that must be fast, cheap and predictable: triage, routing, scoring, guardrails.

Quick start

1. Get a backend.
   • Local: install Ollaya, then pull a model, for example ollaya pull laya (a small model that also runs on a CPU). With Docker: docker run -d -p 11435:11435 ghcr.io/ollaya-dev/ollaya.
   • Hosted: create an API key in the TypeSafe console.
2. Install the node in n8n (see Installation).
3. Create an Ollaya API credential, set Base URL and API Key (see Credentials).
4. Add the Ollaya node to a workflow, pick a model, and choose Evaluate or Route.

Installation

On a self-hosted n8n, go to Settings → Community nodes → Install, enter n8n-nodes-ollaya, accept the risk notice, and install. You need to be the instance owner or an admin. The community nodes guide covers the other installation methods.

Credentials

Add an Ollaya API credential with two fields:

When you save, n8n checks the credential by listing the server's models (GET /v1/models).

Good to know

• The API Key field cannot be left empty. If your Ollaya server runs without OLLAYA_API_KEY, type any placeholder text.
• If n8n runs in Docker, localhost means the n8n container itself, not your computer. Use an address the container can reach (for example http://host.docker.internal:11435 with Docker Desktop, or the server's LAN address).
• Your server must expose GET /v1/models and POST /v1/systemone, and accept Authorization: Bearer <key>. Ollaya and TypeSafe AI both do.

Operations

Both operations send one request per input item, containing the item's state, the selected Model and your questions. All the questions of a request are asked about the same state.

State Format chooses where the state comes from:

Model lists the models available on your server, so with Ollaya you see the models you have pulled. How to structure a state is covered in the TypeSafe docs, and the Ollaya model catalogue lists what you can run locally.

Evaluate

Evaluate asks one or more questions and adds every answer to the item. There are three question types:

Build the questions with the node's fields, or set Questions Format to Using Raw JSON to pass a JSON object in the API's own format, for instance one produced by an earlier node. See the API reference.

Answers land in answers, keyed by each question's ID:

{
  "answers": {
    "is_urgent":   { "noul": 0.95 },
    "department":  { "choice": "billing", "confidence": 0.81 },
    "frustration": { "score": 1.05, "confidence": 0.92 }
  },
  "model": "jev-1.13.0"
}

When an AI Agent uses the node as a tool, it runs Evaluate.

Route

Route asks a single question and sends each item to the output that matches the answer. The Question Type decides which outputs you get:

Each type has its own way to handle doubt:

• Choice can add a Fallback output. Set Confidence Handling to Route to Separate Fallback Output, and any item whose confidence is under Confidence Threshold leaves there.
• Noul starts with both thresholds at 0.5, so everything goes to True or False. Move them apart (for example 0.8 and 0.2) to get an Uncertain output for the answers in between.
• Score levels are numbered from 0, lowest first. The boundary between two levels sits halfway between their numbers: a score from 0.5 up to, but excluding, 1.5 goes to level 1, and a score exactly on a boundary goes to the higher level.

The item gets a route field holding the answer, in the same shape as Evaluate:

{ "route": { "choice": "billing", "confidence": 0.81 }, "model": "jev-1.13.0" }

Example workflow

A support-ticket triage. A webhook receives each ticket and the node's Route operation asks a Choice question with three routes: billing, tech support and sales, each leading to the right team. Confidence Handling is set to a separate fallback output, so a ticket the model is unsure about leaves from Fallback and a Switch node sends it either to a review flag or to a human queue.

Output

Three settings under Options shape the call and the result:

• Simplify (on by default) keeps only noul, choice or score for each answer, plus confidence where the type has one. Turn it off to get the full answer object, including probabilities, and a usage field.
• Include Other Input Fields (on by default) keeps the incoming item's fields and binary data and writes the node's fields on top, replacing any field with the same name. Off, the item contains only the node's fields.
• Timeout is how long, in milliseconds, the node waits for the server to start answering. Default 5000, minimum 1000. A large model on a CPU can need more.

Errors

On failure the node stops and shows the HTTP status together with the server's message. A configuration problem, such as empty Instructions or two routes with the same name, tells you what to fix.

• Turn on Retry On Fail in the node's Settings to retry failed requests, rate limits (429) included.
• Set On Error to keep the workflow going:
  • Continue (using error output) adds an error output that receives each failed item with an error field.
  • Continue sends the failed item, with an error field, to a regular output:

Typical connection problems: a wrong Base URL, a server that is not running, or n8n in Docker pointing at localhost (see Credentials).

Compatibility

Written against n8n 2.40. It also installs on 2.37. Requires a backend that implements the TypeSafe System One API, such as Ollaya or the TypeSafe AI cloud.

Credits and licence

• This node: MIT, see LICENSE.md. Based on the TypeSafe AI community node, whose copyright notice is kept in the licence file.
• Ollaya: the local decision-model server, created and maintained by its own authors under the Apache-2.0 licence. Each model it serves keeps its own licence.
• TypeSafe AI: the hosted System One API and the Jev models.

Resources

• Ollaya project · website · model catalogue
• TypeSafe AI documentation · API reference
• n8n community nodes documentation
• Report an issue with this node

Version history

See CHANGELOG.md.