Data, privacy and memory
For the Personal Dashboard Kit v1.0 (downloaded edition) · checked against the source code and tests on 19 October 2026
"Memory" can mean several different things. This page goes through each place your information could be, one at a time.
Summary
| Your information | dashboard.html |
dashboard-ai.html |
|---|---|---|
| Your spreadsheet | Read in the browser tab; never uploaded | Same. The file itself is never sent anywhere |
| Data sent to an AI provider | None (connections are blocked) | A summary of the current view, only when you click Run agent; column names and values (no rows) when you click Suggest a view |
| Your API key | n/a | In tab memory only; sent only to your chosen provider |
| Saved on your computer by the dashboard | Nothing (except files you choose to download: PDFs, saved agents) | Same |
| Sent to Fractional Partners Asia | Nothing. The kit has no connection to FPA servers | Nothing |
| Persistent memory, chat history, run history | Not implemented | Not implemented |
Layer A: what is sent to the AI for one run (temporary AI context)
When you click Run agent, one request goes to the provider address shown in the preview. It contains:
| Included | Details |
|---|---|
| System instructions | Fixed rules: use only the figures given, don't invent numbers, plain English, under 220 words |
| The agent's instructions | The built-in, pack or custom agent's task text. For Ask your own question, your question |
| A summary of the current view | Number of rows; active filters; the measure shown; the list of allowed columns and their types; the key-number cards; the measure by month (last 24 months); for each allowed category column, the top 12 values with their totals and counts |
| 20 example rows | Only if you choose Summary + 20 example rows. Allowed columns only |
| The file name | Only if you tick Include the file name (off by default) |
| Your API key | In the request header (x-api-key for Claude, Authorization: Bearer for others). Never in the message or address |
| Model name and limits | E.g. max_tokens: 1200 for Claude |
Not included: - your file - other tabs of your workbook - earlier questions or answers - hidden columns - the licence details in your copy - anything from other agents
Data minimisation (implemented and tested). These columns are never sent, and their checkbox is greyed out: - columns the dashboard treats as IDs or free text - columns whose heading suggests a person or contact detail: name, email, phone, mobile, contact, NRIC/FIN, passport, address, postal, date of birth - columns headed exactly Staff, Employee, Manager, Customer, Client, User, Agent, etc.
You can untick any other column. If a filter is on a hidden column, the summary says only "a filter on a hidden column", not its value.
What can still get through. These rules work on column headings. A column with an unhelpful heading, such as "Column F" or "PIC", could still hold people's names, and its top values would be sent. Always check the preview and untick anything personal.
The preview is accurate. The dashboard builds the preview and the real request with the same function. Automated tests confirm the preview text equals the system instructions plus the message actually sent, for both Claude and OpenAI formats. The only parts not displayed are the API key and technical settings (model name, response length limit).
Once sent, the request is handled by the provider (Layer D). The dashboard can't recall or delete it.
Set up the view for me: what it sends
When you click Suggest a view, one request goes to the same provider. It contains: - your request - the names and types of the columns the agent may see - for each allowed category column (up to 30 values), the list of its values, so the AI can choose a filter - the period options - the measure and chart-group names
It does not include rows, totals, hidden columns or the file name. You can see the exact message under What this sends. Nothing in the dashboard changes until you click Apply.
Layer B: browser storage
Checked in the source code (there are no calls to these storage features) and by an automated test (all empty after use):
| Mechanism | Used? |
|---|---|
| localStorage, sessionStorage, IndexedDB, cookies | No |
| Tab memory (JavaScript variables) | Yes, for: your loaded data, API key, provider and model settings, column choices, custom agents, imported packs, the last answer |
What survives what:
| Action | Kept? |
|---|---|
| Refresh / close tab / restart browser | Nothing is kept. You re-open your file, re-enter the key, and re-import any saved agents |
| Switch device | Nothing carries over |
| Files you download (Save my agents, Print/PDF) | Stay where you save them, until you delete them |
Browser risks. - While the tab is open, the key is in the page's memory. - Anyone with access to your unlocked computer, a malicious browser extension, or malware could read it. - Use your own device, keep it locked, review your extensions, and use a key you can revoke. - The key box is a password field, so it shows dots. Browsers don't normally save it, but some password managers may offer to; decline.
Layer C: application backend and database
The downloaded kit has no backend.
- Verified in code: dashboard.html has a safety lock (connect-src 'none') that blocks every network connection. Fonts are built into the file.
- Verified in code: dashboard-ai.html can connect only to https://api.anthropic.com, https://api.openai.com, http://localhost, http://127.0.0.1, and any address you add yourself.
- Verified in code: neither file contains analytics, error-reporting or tracking code.
- Tested: the offline edition made zero network requests in an automated test, even when code tried to connect.
What FPA does hold relates to buying the kit, not to your dashboard data: - the order form details you submit on fractionalpartners.asia (name, email, company, phone, message) - invoice and receipt records - your licence record (name, company, licence number, issue date, download count)
These are kept in FPA's admin systems for accounting and licensing, under FPA's Privacy Notice.
Hosted plan. This page covers the downloaded kit. The hosted plan, on an FPA web address, comes with its own data-handling notice, given to you before your address is set up.
Layer D: storage by the AI provider
Sending a request means that provider's terms, settings and retention rules apply. The dashboard has no control over this. Check the provider's current documents; they change.
| Provider | What their documentation said when checked (19 Oct 2026) | Source |
|---|---|---|
| OpenAI API | API data is not used for training by default. Abuse-monitoring logs may be kept for up to 30 days unless longer retention is legally required. Zero Data Retention is only for eligible, approved customers. | OpenAI: Your data |
| Anthropic API | Retention depends on the model, features and your agreement. The Messages API (used by this kit) is eligible for Zero Data Retention by arrangement. Some models and features have their own retention, e.g. 30 days for certain covered models. Flagged content may be kept longer, and data may be kept where the law requires. | Anthropic: API and data retention, Privacy Center |
| Company or local AI | Depends entirely on that server. It may log requests, or pass them on to another AI provider. | Ask the server owner |
| Gateways or proxies in between | Their own policies apply | Ask your IT team |
No provider is promised to keep nothing.
Persistent memory, history and knowledge bases
| Concept | What it means | In this kit |
|---|---|---|
| Agent instructions | The text that defines an agent's job | Yes, in tab memory, and in files you save |
| Agent configuration | Settings needed to recreate an agent | Name and instructions only; provider and model are set separately and never saved |
| Current request context | Data and instructions for one run | Yes (Layer A); built fresh for every run |
| Conversation history | Earlier questions and answers | Not implemented. Each run is a single message, verified by test |
| Execution history | Records of runs | Not implemented. A new answer replaces the old one on screen; nothing is logged |
| Persistent memory | Information saved for future runs | Not implemented |
| Knowledge base | Documents an agent can look things up in | Not implemented |
| Provider-side retention | What the provider keeps | Depends on the provider (Layer D) |
Agents do not have persistent memory. Having instructions, or being able to read the current dashboard, is not memory.
Deleting your information
| What | How |
|---|---|
| Data, key, agents in the dashboard | Close or refresh the tab. Forget key removes the key immediately. Clear data removes the loaded file |
| Your own agents | Select one → Edit this agent → Delete this agent. Delete any saved .agents.json files yourself |
| Downloaded PDFs and saved agent files | Delete them from your computer |
| Data already sent to an AI provider | The dashboard can't delete it. Use the provider's own controls or contact them |
| Your purchase and licence records at FPA | Contact hello@fractionalpartners.asia. Some records must be kept for tax and company law (usually 5 years in Singapore) |
Your responsibility when using AI
You are responsible for ensuring that you have the necessary rights and permissions to submit data to the AI provider or service you select. Assess the sensitivity of the data, your organisation's policies, and any applicable legal or contractual requirements before using AI features.
The availability of an AI provider or local endpoint does not, by itself, establish that a particular use is secure, compliant, or appropriate.
Practical rules - Only use data you're allowed to process, and only the columns the question needs. - Don't put passwords, keys, tokens or unrelated personal details into data, questions or instructions. - Follow your organisation's data classification and AI policies, and get approval before sending confidential business information to an external AI service. - Check the provider's data terms before using sensitive information. - Review answers before using or sharing them. - Singapore: if the summary could include personal data (anything that identifies a person), the Personal Data Protection Act 2012 (PDPA) may apply. That covers purpose, consent, protection and transfers outside Singapore. The kit is not PDPA-certified and does not make your use compliant by itself. This page isn't legal advice.
Also in this guide set: