Nuve AI

Overview

Nuve AI ("Nuve Guru") is a Claude-based AI assistant for BOM analysis, engineering questions and specification/requirement extraction. It works as a floating chat panel that can be opened from anywhere in the application, and the moment it opens it pulls a live status summary (context) from the BOM Intelligence, Change Management (ECR/ECO/ECN), Requirements, DPP and PDM Vault modules — so it is not a general-purpose chatbot; it answers against the real data in your current session.

Key Concepts

• Active analysis context — every time the panel opens, a live cross-module summary is gathered (/api/ai/guru/context): open ECR/ECO counts, requirement traceability coverage, DPP alerts, PDM Vault file/lock counts. The assistant talks using this data rather than describing a generic product's interface.
• Guru action — a card the assistant proposes instead of changing data directly, waiting for you to approve it. It can only propose two things: opening a new task (propose_create_task) or moving an issue to "Linked to Action"/"In Progress" (propose_patch_issue_status). The model can never close an issue as "Resolved" or "Accepted" — that decision belongs to a human.
• BOM Expert (Nuve Chat) — a separate, narrowly scoped chat assistant inside the BOM Intelligence screen that talks to the currently loaded BOM table and, if connected, the ERP product list.
• Provider chain — answers are produced through Anthropic (Claude) → Groq (free tier, OpenAI compatible) → a rule-based fallback, in that order of priority; even with no API key the assistant returns an answer computed from real analysis data rather than going blank.

Typical Workflow

1. Open the chat launcher at the bottom right of any screen (Nuve AI).
2. Ask a question (e.g. "What are the 3 most critical issues?", "Why is our domestic content ratio low?", "What is the impact of this ECO?") — the assistant answers using the active analysis context.
3. If needed, select an action target (an issue) and ask the assistant to open a task or update the issue status.
4. The assistant shows an action proposal card (with preview text); the decision stays with you via Approve or Cancel.
5. An approved action is applied through the same service function the human interface uses, and under your own role/permissions.
6. To extract requirement candidates from specification text, use the AI Extraction flow (Kickoff Pack / Spec2Action context).

Feature Reference

• Cross-module live context gathering (/api/ai/guru/context): Change Management, Action Hub, Requirements, DPP alerts, PDM Vault
• Chat (/api/ai/guru/chat) — produces answers from the conversation history plus the live context
• Action proposal flow: proposal creation, approval (/api/ai/guru/actions/{id}/confirm) and cancellation (/api/ai/guru/actions/{id}/cancel) — proposals expire by themselves after 10 minutes
• BOM Expert chat (/api/ai/bom-expert/chat) — specific to BOM Intelligence, limited to the current BOM/ERP data
• Specification extraction (/api/ai/specification/extract) and requirement rewriting (/api/ai/requirements/rewrite)
• Action proposals generated from issues (/api/ai/issues/recommend-actions)
• Kickoff pack creation (/api/ai/kickoff-package): extracts requirement candidates from specification text and turns them into a starter package
• Anthropic → Groq → rule-based fallback provider chain; works without an API key

Constraints / Known Limits

• The assistant can only propose two actions, both requiring human approval and both reversible; closing issues, ECR/ECO/ECN operations and cost simulation are deliberately out of scope — those decisions require a multi-step approval chain.
• If no API key is configured, answers fall back to a rule-based (deterministic) path; in that case predefined patterns are used rather than free-text generation.
• Answers are suggestions; for critical decisions (closing a status, approving a change) they are not applied without human verification.