AI Features

On the New Case page, click AI Draft and paste raw notes, error-log excerpts, shift-handover messages, or a brief description. Claude structures the text into three clean sections:

• Observation: what was seen, heard, or measured, with timestamps where present.
• Root cause: the likely or confirmed cause, if it is identifiable from the input.
• Resolution: the steps that fixed the issue, or what was attempted.

Best practices

• Paste as much raw context as possible — Claude does better with more material than less.
• Include timestamps, error codes, and equipment IDs verbatim.
• Always review and edit the draft. The AI is a starting point, not a verdict.
• If you don't have a root cause or resolution yet, leave those notes as 'TBD' and the AI will keep the fields empty.

What it will NOT do

• Invent severity levels. You must select severity yourself.
• Pick a workspace or category. These remain manual selections.
• Fabricate technical details that are not present in your input.
• Reference cases or documents from outside your organization.

The blue chat bubble in the bottom-right corner (the FTS-Clanker mascot) opens the AI Copilot. The Copilot has access to all verified cases in your organization and can answer natural-language questions.

Example prompts

How it works

Each question triggers a semantic search across your verified cases. The top five most-relevant cases are passed to Claude as context, together with your question. Claude composes an answer that cites specific case IDs in square brackets such as [#42, #117]. Click any cited ID to open the full case.

Honesty guarantee

If no relevant verified case exists, the Copilot says so explicitly: "I don't have any verified cases matching this question. Try searching with broader terms, or create a new case if this is a fresh issue." It never invents fictional case IDs or fabricates resolution steps.

On any case detail page, click the Translate button to get a full DE ↔ English translation of the observation, root-cause, and resolution fields. The original text remains unchanged; the translation appears in a side-by-side panel.

Typical use cases

• Cross-border teams in the DACH region (Germany, Austria, Switzerland).
• International suppliers reading service-bulletin equivalents.
• Onboarding new staff with different language backgrounds.
• Audit reports for English-speaking auditors of German operations.

Translation preserves technical vocabulary, error codes, and equipment names verbatim. Domain-specific terminology (for example "Spindel-Vibration" or "Kühlmitteltemperatur") stays consistent across cases thanks to context-aware glossary handling.

When viewing a case, the right-hand sidebar shows Similar Cases— up to five cases ranked by semantic similarity to the current case's observation and tags. Verified cases appear first, marked with a green badge.

Ranking signals

• Text similarity (cosine distance on embeddings) of the observation field.
• Tag overlap (Jaccard similarity).
• Same workspace > same site > same organization (decreasing weight).
• Verified status (verified cases get a boost).
• Recency (a lighter boost for cases from the last 90 days).

Operators routinely check Similar Cases before submitting. This prevents duplicates and helps surface known resolutions. Reviewers use it to identify cases that should be merged or that point to recurring root causes.

FTS treats AI privacy seriously. Three commitments:

• No training on your data: Anthropic's API terms — which FTS contractually binds itself to — prohibit using customer prompts to train models. Your cases stay your cases.
• Per-organization isolation: each Copilot query is scoped to your organization's verified cases only. Claude cannot read across organizations.
• Fully auditable: every AI request (draft, chat, translate) is logged in the audit trail with timestamp, user, and request ID. Admins see usage statistics from Settings → Audit → AI Usage.

FTS uses different Claude models depending on the task:

• Claude Sonnet 4.6: the primary model for Copilot chat and case drafting. It offers the best balance of quality and speed.
• Claude Haiku 4.5: used for fast similarity ranking and short translations to keep latency under one second.
• Claude Opus 4.6: an Enterprise-only option for complex multi-case synthesis (for example "summarize all spindle-vibration cases from the last year and identify trends").

Model selection is automatic per task — you cannot manually choose a model. We monitor latency and quality continuously and may swap models as Anthropic releases new versions, always preserving backward-compatible behavior.

To ensure fair usage and protect against runaway costs, AI features are rate-limited per user:

The Starter plan requires the AI add-on (€49 per month) to enable any AI feature. When you hit a limit, the UI shows a friendly cooldown message and no requests are lost.

The Copilot says "No matching cases"

This means your organization has no verified cases matching your query. Drafts and rejected cases are excluded. Either broaden your question or verify some draft cases first.

AI Draft fills the wrong workspace

The AI does not auto-select a workspace — you do. If you see a workspace already filled in, it was inherited from your last new-case session, not from the AI.

Translation looks off

For domain-specific terms that are not in the default glossary, ask your Admin to add them to Settings → AI → Glossary (an Enterprise feature). For common technical German ↔ English, we rely on Claude's native multilingual capability.

• FAQ — common questions about plans, privacy, and integrations →
• Admin Guide — billing, taxonomy, and audit log →
• Getting Started — set up your first case →