Nucleus overview - Intellikon AI

Nucleus is where your team writes and looks after your workspace’s AI prompts. A prompt is the instruction an AI model works from, such as “read this ticket and give it a category and a priority”. In Nucleus you keep every version of a prompt, try a change before anyone relies on it, test versions against sample inputs, and release the version you trust. You can then see how each prompt performs and what it costs. Nucleus holds your workspace’s own prompts. It does not change the built-in prompts behind Studio’s own features. It is also not where you decide which AI models your workspace may use: Tenant Owners and Workspace Admins do that in Studio settings, under AI Models. Nucleus is for the people who build and look after your AI. Tenant Owners and Workspace Admins can do everything in it. Workspace Members work on prompts, datasets and evaluations day to day.

How Nucleus is laid out

The top of the Nucleus Dashboard, with Total prompts, Healthy, At risk and Workspace resolution, and the first prompts with their health scores.

Area	What you do there
Dashboard	See a health score for each of your active prompts.
Prompts	Write prompts, save versions and release them.
Datasets	Keep sets of sample inputs to test prompts with.
Evaluations	Run prompts against inputs and score the answers.
Analytics	See runs, tokens and cost from the playground and evaluations over the last 30 days.
Health	Get alerts when a prompt’s scores start to slip.
Observability	Look at individual AI requests, with the model, time taken and cost.
API Keys	Create keys that let your own software use your released prompts.
Inbox	Read notifications about your prompts.
Settings	Set up the products and types you file prompts under.

Key ideas

Prompt

An instruction for an AI model, filed under a product and a type.

Version

Each save that changes the prompt, model or settings adds a version, and the versions before it stay in the history.

Environment

dev, staging or production, each holding one version of a prompt at a time.

Dataset

A set of sample inputs to test prompts against.

Evaluation

A run of a prompt against inputs, with the answers scored by any scorers you add.

Scorer

The marking guide an AI judge uses to score an answer.

From draft to production

You write a prompt and save a new version each time you improve it. When a version is ready, you promote it to an environment. Before it moves, Nucleus checks the version’s latest automatic evaluation. A version that passed goes straight through. Otherwise, what happens depends on the prompt’s quality gate: Nucleus lets it through, asks you to say why it should go ahead, or stops it unless a Tenant Owner or Workspace Admin gives a reason.

Who can do what

Task	Tenant Owner	Workspace Admin	Workspace Member
Open Dashboard, Prompts, Datasets, Evaluations, Health and Observability	Yes	Yes	Yes
Create a prompt	Yes	Yes	No
Edit a prompt and save versions	Yes	Yes	Yes
Go back to an earlier version, or turn a prompt off and on	Yes	Yes	No
Promote a version to dev or staging	Yes	Yes	Yes
Promote a version to production	Yes	Yes	No
Set a prompt’s release policy	Yes	Yes	No
Create datasets and run evaluations	Yes	Yes	Yes
Create and change scorers	Yes	Yes	No
Open Analytics	Yes	Yes	No
Change Settings	Yes	Yes	No
Create API keys	Yes	Yes	No
Open Inbox	Yes	Yes	Yes

Where to start

Write a prompt

Create a prompt, try it and save new versions.

Release a version

Promote a version to dev, staging or production.

Test your prompts

Build datasets, run evaluations and score the answers.

Watch how they perform

Check health scores, alerts, requests and cost.

Tenant Owners and Workspace Admins set up the products and types prompts are filed under in Nucleus settings.

Vera

Chatbots that answer from your own knowledge.

Get started

Move between products and find Studio settings.

​How Nucleus is laid out

​Key ideas