Personal Dashboard Kit · Guide

AI setup guide

For dashboard-ai.html · last checked 19 October 2026

This guide assumes you have never set up an AI API before.


Step 1: Choose the edition

  • dashboard.html has no AI. It cannot connect to the internet at all. Nothing to set up.
  • dashboard-ai.html is the same dashboard plus AI agents. It needs an AI provider (Step 2) before agents will run. Everything else works without one.

Step 2: Choose an AI provider

Open dashboard-ai.html, load a file (or click Try sample data), then click Set up connection in the AI agents panel. The AI provider list has three options:

Option Choose it when
Claude (Anthropic) You have, or can create, an Anthropic API account
OpenAI You have, or can create, an OpenAI API account
Company or local AI (OpenAI-compatible) Your organisation provides an approved AI server, or you run a model on your own computer (for example with Ollama or LM Studio)

Pick whichever your organisation allows. The kit doesn't rank providers on quality, privacy or security; check your organisation's policy and the provider's own terms.

A subscription is not API access. A Claude.ai plan (Pro, Max, Team) or a ChatGPT plan (Plus, Pro, Team) does not include API usage. The API is billed separately, per use, in the provider's developer console. See Anthropic's explanation and OpenAI's explanation.

Step 3: Get and enter your API key

⚠ Your API key is a credential that can authorise usage against your provider account. Keep it private. Do not include it in screenshots, shared configuration files, public repositories, or support requests, including emails to us. We will never ask for it.

Claude (Anthropic) 1. Sign in at the Anthropic Console and add billing or credit (how to get an API key). 2. Create a key. Give it a name like "Dashboard kit, my laptop". 3. Copy it once and paste it into the API key box. You may not be able to view it again in the console.

OpenAI 1. Sign in at platform.openai.com, add billing, and open API keys. 2. Create a secret key. If you can, give it access only to the model endpoints. 3. Paste it into the API key box.

Company or local AI 1. Address: enter the server's base address, for example http://localhost:11434/v1, or the address your IT team gives you. If you paste a full address ending in /chat/completions, that's fine; the dashboard handles it. 2. API key: enter one only if the server needs one.

What the dashboard does with your key - It is kept in the open browser tab's memory only. It is never written to disk, browser storage, the agent files you save, or the preview. - It is sent only in the request header to the provider address you chose. - Click Forget key to remove it. Closing or refreshing the tab also forgets it. - You'll paste it again each time you open the dashboard. This is deliberate.

Step 4: Enter the model name

Type the model identifier exactly as your provider lists it. - Claude: the box is pre-filled with a current model name. Check Anthropic's model list. - OpenAI: check OpenAI's model list, or your account's model page. - Company or local AI: ask the server owner, or use the name of a model you have installed locally.

Model names, prices, limits and availability change over time, and your account may not have access to every model. Larger models usually cost more per use.

Step 5: Test the connection

Click Test connection. It sends one short message ("Reply with OK") and none of your data.

  • Success looks like: ✓ Connected to Claude (Anthropic) · model …
  • For OpenAI, the test first checks your key and confirms that your account has the model name you typed, then sends the short message.

If something goes wrong. API keys are never shown in these messages, not even partly.

What you see What it means What to do
The API key was refused Wrong, expired or revoked key, or a key from the other provider Copy the key again in full; create a new one if needed
The model name wasn't recognised / no model called … Typo, retired model, or your account can't use it Copy the exact name from the provider's model list
Out of credit or has hit a usage limit No credit, spending limit reached, or too many requests Check billing and limits in the provider console; wait and retry
The AI provider had a problem on its side Provider outage or overload Try again later
Couldn't get a readable answer from OpenAI Often a refused key (OpenAI doesn't let browsers read chat error messages), or no internet Run Test connection, which reads the error properly
Couldn't reach … / allow requests from a browser No internet, server down, or a company/local server that blocks browser requests (CORS) Check the server is running. For Ollama, start it with OLLAMA_ORIGINS=* (or list your origins)
The safety lock in this file doesn't allow … The company server's address isn't on the allowed list See "Adding a company AI server" below
Only addresses on this computer may use http:// A company server address starts with http:// Use its https:// address
Not in the expected format / no text The server isn't fully OpenAI-compatible Ask the server owner for an OpenAI Chat Completions endpoint

Adding a company AI server to the safety lock

dashboard-ai.html can only connect to the addresses listed in its safety lock. By default these are Anthropic, OpenAI, and localhost / 127.0.0.1 on your own computer. To add your organisation's server:

  1. Open dashboard-ai.html in a text editor (Notepad, TextEdit in plain-text mode).
  2. Near the top, find the line containing connect-src.
  3. Add your server's address after the others, separated by a space, e.g. https://ai.yourcompany.com, then save.
  4. Reopen the file in your browser. The dashboard reads the same list, so there is nothing else to change.

Requirements for the server - It uses HTTPS. - It accepts POST {address}/chat/completions in the OpenAI Chat Completions format. - It returns choices[0].message.content. - It allows browser requests from a local file (CORS). - If it needs a key, it accepts it as Authorization: Bearer <key>. Custom headers aren't supported.

Only add servers your organisation has approved. A "local" or "company" server is not automatically private: it may log requests or pass them on to another AI provider. Check with its owner.

Step 6: Run a built-in agent

  1. Open your file and set the filters, period and "Show" measure you want. The agent sees only the current view.
  2. Click an agent, e.g. Explain this view.
  3. Optional: open Choose which columns the agent may see and untick anything it doesn't need. Columns that look like names, emails, phone numbers, IDs or free text are never sent. The file name is off by default.
  4. Open Preview exactly what will be sent. The text shown is exactly the message sent to the provider. It includes: - the system instructions - the agent's task - a summary that the dashboard calculated: key numbers, the monthly trend, the top 12 values per category, and 20 example rows only if you choose Summary + 20 example rows
  5. Click Run agent.
  6. Check the answer against the charts. - Numbers in the summary were calculated by the dashboard. - Everything else is the AI's interpretation, and can be wrong.

The first time you send a key to a company or local address, the dashboard asks you to confirm.

Step 7: Let the AI set up the view (optional)

In the AI panel, under Set up the view for me: 1. Describe what you want to see, e.g. "Online sales in the East for the last 3 months by product category". 2. Click Suggest a view. The AI replies with a list of suggested changes: filters, period, what to show, and the chart. Anything that isn't possible with the built-in settings is listed separately. 3. Click Apply to use it, or Dismiss. Undo puts the previous view back.

What it can't do. It only chooses settings that already exist. It can't add target lines, new chart types, formulas, colours, layout or data. For those, use the prompts in "PROMPTS - copy and paste into Claude.txt" with your own AI assistant, and keep a copy of the original file.

Step 8: Create your own agent

  1. Click + Create your own agent.
  2. Fill in the two fields that exist: - Agent name: up to 40 characters. - Instructions: up to 2,000 characters. Write it like a brief to a colleague: what to look at, what to produce, how long. You can start from one of the three examples.
  3. Click Save agent. It appears as a button. Select it, check the preview, and click Run agent.
  4. To manage agents: - Your own agents: select one and click Edit this agent to change or delete it. - Built-in or pack agents: click Copy and edit this agent to make your own version.
  5. Your agents disappear when you close or refresh the tab. Click Save my agents to download them as my-agents.agents.json, and use Import agent pack next time. The file contains only names and instructions.

There are no other agent settings. Every agent uses the provider and model in the setup panel. Agents have no tools, no memory and no permissions to configure, because they can't do anything except write text.

Good agent instructions: - Say what data the agent should use, e.g. "use only the breakdowns given". - Ask it to separate facts (numbers) from suggestions. - Ask it to say plainly when something can't be told from the data. - Never put passwords, API keys or personal details in instructions. The dashboard refuses text that looks like an API key.

See AI-AGENT-SAFETY.md for more.


Also in this guide set:

WhatsApp