Provider and models
Artica’s AI features use Anthropic’s Claude models (currently the Claude Sonnet family), through Anthropic’s API, called from our servers. No other AI company receives your content. We may move to a newer Claude model when one becomes available; we will update this page if we ever add another AI provider.
What each tool sends
| Tool | Sent to Claude | Before sending | What comes back | What we keep |
|---|---|---|---|---|
| Article generator | Your capture’s screenshots (up to a set number and size), each step’s click description, the page titles and addresses (site and path only), and where the recording started | Screenshots and text are redacted in your browser for emails, card numbers, tokens, phone numbers, Social Security numbers and IBANs; text is redacted again on our server. Names and other text on screen are not blurred | A draft article: title, summary, steps and a line on how to check it worked | The draft article (which you then edit), the screenshots and steps, and the AI usage and cost |
| Topic finder | Each ticket’s subject and first message, up to 2,000 characters; while reading a file, its column names and a few sample rows, or document lines; then the AI-written questions, to name topics and categories | Personal data is removed first on our processing service: names, companies, emails, phone numbers, card and account numbers, addresses and signatures become placeholders | One anonymous question per ticket; topic and category names and descriptions | Topics, categories, descriptions, example questions (checked not to repeat real ticket text), ticket IDs and counts, plus AI usage and cost. Never ticket text |
| Format fixer (AI fix only) | One article’s cleaned text and structure, its title, and the article type you chose | In your browser, personal data our scrubber finds (emails, phone numbers, card and account numbers, keys and similar) becomes placeholders, and links and images become short references; our server scrubs again | The article reshaped into the template for its type | Counts only (type, issues, success, AI tokens and cost). The fixed article stays in your browser |
Tools that use no AI
Ticket exporter, Article exporter, the Format fixer Check, Pre-built themes and Custom themes send nothing to an AI provider. A raw (“Keep everything as it is”) ticket export is never sent to AI.
Models we run ourselves
Two jobs use open-source models that run on our own processing service (Modal), not at an AI company: finding names, companies and addresses when scrubbing tickets, and grouping Topic finder’s questions by meaning. They run in memory for the job and keep nothing.
No training on your data
Under Anthropic’s commercial terms, content we send through the API is not used to train its models. Artica doesn’t train or fine-tune any model on your content either.
Retention
We use Anthropic’s standard API terms. Anthropic keeps API inputs and outputs for a limited period (currently up to 30 days), and longer where needed to enforce its usage policies or where the law requires. We have not yet arranged zero data retention with Anthropic; when we do, we will update this page. What Artica itself keeps is listed in the table above and in our Privacy Policy.
Limits of AI output
- AI output can be wrong, incomplete or invented, even when it reads well. A drafted article may misdescribe a step; a topic may group tickets that don’t belong together or miss a small topic; a fixed article may drop or change something.
- Format fixer compares the article before and after and tells you if anything looks lost, and Topic finder shows how many tickets each topic covers. These checks help; they don’t guarantee accuracy.
- Scrubbing is done as thoroughly as possible, but no scrubber catches everything.
- Review everything before you publish or rely on it. Our Terms say more.
Human oversight
Artica’s AI drafts and summarises; you decide. Nothing is published automatically: drafted articles go to your Zendesk only as drafts, and only when you export; fixed articles are saved only when you click Save or Replace. Artica makes no automated decisions about people, and nothing it does has legal or similarly significant effects on anyone. Questions: hello@artica.tools.