AI Chat

TODO (New Screenshot): AI Chat panel (App 2.0) Include: session list, prompt template entry, context Q&A area, and copy/export actions. Suggested filename: ai-chat-panel-v2-en.png

Scope

AI Chat handles semantic processing after transcription:

• Conversation history and search
• Multi-turn Q&A
• Prompt library support
• Context-aware summarization/extraction

It does not perform speech recognition itself.

Use Cases

• Meeting summary generation
• Action-item extraction
• Translation and rewrite assistance

Steps

1. Open AI Chat from navigation or Note side panel.
2. Start a new chat and ask a task-specific question.
3. Use prompt templates for repeatable workflows.
4. Iterate with follow-up questions until output is usable.
5. Copy validated output into notes or downstream tools.

Prompt Quality Tips

• Specify target structure explicitly (for example: title + bullets + action items).
• Set boundaries explicitly (for example: “use only transcript evidence”).
• Split long objectives into rounds: summarize → extract actions → draft final output.
• Ask for evidence back-links before accepting critical conclusions.

Term Explanations

• Context: transcript + conversation state available to the current session.
• Prompt template: reusable query structure for consistency and speed.
• Multi-turn workflow: iterative questioning to converge on production-ready output.

Example Workflow (Meeting Minutes)

1. First prompt: Summarize this meeting by agenda topics.
2. Second prompt: Extract action items grouped by owner.
3. Third prompt: Draft a send-ready follow-up email.

Real Scenario: Weekly Project Review

For weekly project reviews, break one “big ask” into three smaller rounds:

1. Constrain facts first: List decisions only, no suggestions.
2. Structure execution next: Rewrite as owner + deadline + risk.
3. Apply tone last: Rewrite for leadership update style.

This sequence improves controllability: each round is verifiable, and failures are easier to rollback.

FAQ

Q: Why does answer quality vary?
A: Provider selection, context size, prompt quality, and quota state all affect output.

Q: Can I use custom AI services?
A: Supported providers and modes are configured in Settings.

Q: Is chat shared across all workspaces?
A: Workspace scoping applies by default to avoid project mixing.

Common Mistakes

• Mistake: prompts too vague (for example: “summarize this”)
  Safer: define audience, output format, and length constraints.
• Mistake: shipping AI output as-is
  Safer: verify key claims against transcript evidence before publishing.
• Mistake: using one long chat for all tasks
  Safer: split sessions by objective (summary/action/external copy).

Output Landing Tips

1. Promote key AI results into Note body (avoid leaving only chat history).
2. Manually verify numbers, names, and dates before external sharing.
3. Standardize team templates for recurring workflows.

Limitations

• Status: Stable (non-Beta), while provider-side capabilities can change over time.
• Availability and quotas depend on subscription and provider policy.
• External AI services require network access and credentials.
• Sensitive data handling should follow your compliance/security rules.
• Platform: Windows and macOS both support AI Chat, with differences mainly in proxy/certificate environment setup.