Transparency

How Naina uses AI

Naina and LLM Insights are the parts of Naina Labs that use AI. Every other tool is ordinary code that runs in your browser and doesn't send your files to our servers.

AI modelsClaude, ChatGPT and Gemini (USA)
Used forNaina and LLM Insights
Chats saved by Naina LabsNever
Trains on your dataNo

The AI behind Naina

Naina uses large language models: Claude from Anthropic on its own, or ChatGPT from OpenAI, Gemini from Google and Claude together when you switch on all three models. A large language model reads text and writes text. In Naina there's one chat and one place to drop a file. For a file, each model gets the veiled text with instructions written and tested by Naina Labs, and returns themes, figures and actions. With all three, Claude then combines their answers, keeps what they agree on and shows where they disagree, with each model's view. You can read every model's own answer too.

The AI doesn't receive your file itself, only text taken from it after Privacy Veil has hidden personal details. In Naina it can't take actions, can't contact anyone and doesn't remember you between sessions. Each analysis is a fresh, one-off request, and when it's working on your file it has no access to the web.

Chat without a file works differently. When you chat with Naina without attaching a file, Privacy Veil isn't active: your message is sent as you typed it, so it can include names of people and organisations you're asking about. To answer questions about current events, Claude may search the web; Anthropic runs the search using words from your question, and the answer lists the sources it used. Please don't type personal details about yourself or other people in chat without a file. Web search is limited to a couple of searches per question and a monthly allowance per person.

We use a fast, efficient model. It suits this job well: Naina Labs does the maths itself, so the AI only has to spot the patterns in the figures and explain them in plain English. We don't use preview or restricted models.

Naina is a Naina Labs product. It isn't made or endorsed by Anthropic, OpenAI or Google.

Naina uses credits: 100 per user a month, with a chat message costing 1 credit (3 with all three models) and a file 3 (8 with all three). Using three models costs more because three companies answer. Credits also stop automated misuse and keep Naina affordable for everyone.

Our cautious approach

We send as little as possible. This is what happens when you send a file to Naina:

  1. Nothing is sent to the AI until you press the button

    Loading a spreadsheet, building charts and browsing your data all happen in your browser.

  2. Personal data is hidden first, by default

    Naina Labs checks every column name and value. Columns that look like personal data are dropped completely. Personal details found inside other columns are hidden by Privacy Veil: each one is replaced with a placeholder such as [PERSON_1] or [EMAIL_1], the same value always gets the same placeholder, and the real value is put back on your screen when the answer comes back. The list linking placeholders to real values never leaves your browser. This runs by default, and you can check and change what it hides before you send.

  3. Your file name and sheet name are withheld

    File names often contain people's names, so we don't send them.

  4. Only what the job needs is sent

    For spreadsheets and social media exports, the AI gets column names, totals and pre-calculated breakdowns such as "revenue by shop", plus a few cleaned example rows or posts. For customer comments, documents and translations, it gets the cleaned text itself, because that's what it has to read.

  5. You can see what will be sent

    Every task shows a "What will be sent" panel, with a count of what was hidden and a preview of the text, before you press go. If anything was found, you first see a list of each item so you can hide or keep it, and afterwards a privacy receipt counts what was hidden (counts only, never the details).

  6. We don't save your chats

    Naina Labs doesn't store your files, the text sent, your questions or the answers, not even in your account. Close or reload the page and they're gone. Your account only keeps a count of how many analyses you've used.

What we remove before sending

Naina Labs looks for these in column names and in the values themselves:

Names of peopleCustomer, client, patient and staff columnsEmail addressesPhone numbersHome addresses and postcodesDates of birth and agesNational Insurance, NHS and passport numbersBank account numbers, sort codes and IBANsCard numbersUsernames, passwords and IP addressesHealth details

Automatic detection has limits

A person's name typed into a free-text note, such as "spoke to Jane about the refund", can't always be spotted. Please don't use Naina on health records, payroll, customer contact lists or anything else that's mainly about individual people. The rest of Naina Labs, including the Dashboard, works without sending anything.

What the AI companies do with it

The cleaned text goes to Anthropic (Claude) in the United States, and also to OpenAI (ChatGPT) and Google (Gemini) when you switch on all three models. Under its policies for business customers:

  • It doesn't use the data to train its AI models.
  • Their business API terms say they don't train on it, and delete requests and answers automatically after their misuse checks: up to 30 days for Anthropic and OpenAI, and 55 days for Google.
  • Content that its safety systems flag as breaking its usage policy may be kept for longer, up to 2 years, so it can investigate misuse.

You can read Anthropic's own explanation in its Privacy Center.

LLM Insights

LLM Insights works differently from Naina. Every week it asks ChatGPT (OpenAI), Gemini (Google) and Claude (Anthropic) the prompts you write, with web search on, then a second AI read (Gemini) scores each answer for whether your brand is named, its position, sentiment, themes and claims that contradict your brand facts. Please keep personal details out of prompts and facts. We name the exact model behind every answer. Unlike Naina, LLM Insights stores your prompts and the answers for 6 months so it can show trends.

Naina's readout is narrated by an AI voice from ElevenLabs, and the player says so. ElevenLabs receives only the readout script, which is written from your results, and we save the audio for up to 6 months so it can be replayed. Brand logos come from Brandfetch and are loaded by your browser.

What Naina can and can't do

It can

  • Spot trends over time
  • Compare groups, such as shops or products
  • Point out unusually high or low figures
  • Suggest questions worth looking into
  • Show you every comment behind a finding, so you can check it
  • Flag customer comments that may need a person to look, such as legal threats, safety risks or customers who may be vulnerable

It can't

  • Guarantee every figure is right
  • See data you didn't upload
  • Know the reasons behind your numbers
  • Replace advice from an accountant, doctor or lawyer
  • Give legal, financial or medical advice
  • Create or edit images
  • Work with court documents, police matters or personal legal issues

Naina is told to use only figures from your data, but AI can still make mistakes. Check anything important against your spreadsheet before you act on it.

Flags are prompts, not decisions. When Naina flags a comment, for example because the customer may be vulnerable, it is only pointing it out so it doesn't get missed. It can be wrong in both directions. A person should read each flagged comment and decide what to do, following your own policies.

Appropriate use: Naina is for business data only. Before anything is analysed, it's checked automatically. Court documents and forms, witness statements, police matters, material from legal proceedings and personal legal or criminal matters are refused, the session is cleared and the account gets a strike. After three strikes, access to Naina is removed and the latest payment refunded. Read the appropriate use policy.