> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ferrufino-tech-solutions.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Assistance

> Use AI-powered features powered by Anthropic Claude to draft cases, translate, and answer questions.

FTS integrates Anthropic Claude to accelerate knowledge capture and discovery. What decides whether an AI feature is available to you is your **plan**, not a setting: there is no AI settings area and no per-workspace or per-feature switch — see *Enabling & Disabling AI* below. (Open item #89.)

## Available AI Features

### Draft Cases from Notes

When you have rough notes or incident details, the new-case form can turn them into a structured case:

1. Open **New Case** from the sidebar
2. Type or paste your notes into the **Observation / Symptoms** field (at least 10 characters)
3. Click **AI Assist — structure my notes**
4. Claude fills in the rest of the form:
   * A clear, specific title
   * A structured observation
   * A root cause and a step-by-step resolution, where they can be inferred
   * A severity and a category
   * Three to six suggested tags
5. Review and edit every field before saving

This saves hours of manual writing, especially for incident documentation.

<Note>
  There is no separate upload here. The assistant reads what is in the observation field — nothing else — and it writes into the form you already have open.
</Note>

<Tip>
  The generated draft is a starting point. Always review and add your own insights before submitting for review.
</Tip>

### Translate Cases

Need to share a case with a colleague who reads the other language? Every case carries one translate button:

1. Open any case
2. Click **Translate** — the button names the target language, so a German case offers **English** and an English case offers **German**
3. Claude translates title, observation, root cause and resolution, preserving technical terminology
4. Click the button again to hide the translation

<Warning>
  Two limits worth knowing before you plan around this feature. **English and German are the only two languages** — there is no target-language picker and no third language. And the translation is **shown, not stored**: it appears below the case for as long as you keep it open and is gone when you leave the page. It does not become a second, linked case, and nobody reviews or publishes it.
</Warning>

### Summarize Threads

<Warning>
  **This does not exist yet.** There is no summarise button anywhere in the product, and no **AI Summary** control. What a case does have today is **Case History** in the case view: a dated list of every change, review and evidence upload — unsummarised, and showing the most recent 200 entries. The section below describes what is planned, not what you can click. (Open item #89.)
</Warning>

If a case has extensive review comments or a long revision history, a summary of the key points would let you catch up before re-opening an old case. Today you read the **Case History** entries yourself.

### Answer Questions

Search can only find existing cases. The copilot lets you ask questions in your own words:

1. Click the **FTS-Clanker** bubble in the bottom-right corner of any signed-in page
2. Type your question in natural language (e.g., "What should we do if the hydraulic pump makes noise?")
3. Claude answers, drawing on your cases whose status is **Verified** or **Published** — a case still in Draft or In Review is not offered to it, and neither is a Deprecated one
4. Click the bubble again to close it

<Note>
  The copilot is part of the **Team** plan and above. On the Starter plan the bubble is not shown at all — deliberately, so that you are not offered something that would fail when you used it.
</Note>

This feature is especially powerful for:

* Troubleshooting on the production floor
* Training new technicians
* Quick reference during incidents

<Warning>
  AI answers are based on your Verified and Published cases only. If no cases address your question, AI will say so rather than guess.
</Warning>

## How It Works Technically

All AI features use **Anthropic Claude** (the same model powering this documentation). Here's what happens:

1. Your query or content is sent to Anthropic's servers
2. Claude processes it and returns results
3. Results are displayed in FTS
4. Your data is **not used** to train or improve Claude

Each request is isolated—Claude doesn't retain memory of previous interactions with your workspace.

## Privacy & Data Security

<CardGroup cols={2}>
  <Card title="No Training Data" icon="shield">
    Customer cases and queries are never used to train AI models.
  </Card>

  <Card title="Real-time Processing" icon="zap">
    Requests are processed immediately and not stored by Anthropic.
  </Card>

  <Card title="GDPR Compliant" icon="lock">
    Data handling meets EU data protection standards.
  </Card>

  <Card title="Plan-based availability" icon="settings">
    The copilot is limited to the Team plan and above. There is no per-workspace AI switch — see below.
  </Card>
</CardGroup>

## Enabling & Disabling AI

<Warning>
  **This does not exist yet.** There is no AI settings area and no per-feature switch. What decides whether an AI feature is available today is your **plan**, not a setting: the copilot needs Team or above, and the draft and translate buttons are shown to everyone who can edit a case. If you need AI switched off for a sensitive workspace, write to us and we will discuss it — there is no toggle for you to flip. (Open item #89.)
</Warning>

## Quality & Limitations

AI features work best when:

* Case content is clear and well-written
* Evidence is well-captioned
* Tags accurately describe the case
* Questions are specific and grounded in your knowledge base

AI has limitations:

* **Draft cases** still require human review and editing
* **Translations** need human review for specialized terminology
* **Answers** can only cite cases whose status is Verified or Published—they won't extrapolate beyond your knowledge base

<Accordion title="What if the AI translation is inaccurate?">
  Correct it yourself in the case text, then translate again. **There is no translator role and no translation review step.** FTS has exactly three roles — Operator, Reviewer and Admin (`enum UserRole`) — and nobody can be assigned as a translator. Nor is a translation ever published: as the warning above says, it is shown beside the case and gone when you leave the page. AI is a starting point, not a final product.
</Accordion>

## Feedback & Improvement

If you find an AI-generated draft unhelpful or a translation inaccurate, write to
[hello@ferrufino-tech-solutions.com](mailto:hello@ferrufino-tech-solutions.com) and describe what
could improve.

<Note>
  There is no feedback button inside the product, and nothing you type into an AI feature is forwarded to Anthropic as feedback. Your prompts are sent to Anthropic only to answer that one request, and are not used to train models — see “Privacy & Data Security” above.
</Note>

## Best Practices

<AccordionGroup>
  <Accordion title="Review all AI-generated content">
    Don't publish without careful review. AI can make mistakes.
  </Accordion>

  <Accordion title="Ask specific questions">
    "What's best practice for motor maintenance?" works better than "Tell me about motors."
  </Accordion>

  <Accordion title="Use translations for training">
    AI translations are excellent for new-hire onboarding in multilingual teams.
  </Accordion>

  <Accordion title="Improve your case library">
    The more clear, detailed cases you publish, the better AI answers become.
  </Accordion>
</AccordionGroup>

<CardGroup>
  <Card title="Search" icon="search" href="/en/features/search">
    Combine AI answers with semantic search for powerful discovery.
  </Card>

  <Card title="Cases" icon="file-text" href="/en/features/cases">
    Improve case quality to unlock better AI assistance.
  </Card>
</CardGroup>