Quarterly review
January 30, 2026 · View on GitHub
FAQ
Q: Is my data private with Ollama?
A: Yes! Ollama runs 100% locally on your machine. No data is ever sent to external servers. It's GDPR and HIPAA compliant.
Q: Which AI provider is best?
A:
- Privacy-first: Ollama (local, free)
- Best reasoning: Claude (cloud, $$)
- Most versatile: OpenAI (cloud, $$$)
- Fastest: Groq (cloud, $)
Start with Ollama, upgrade to cloud if needed.
Q: How much does cloud AI cost?
A:
- Claude: ~$0.02 per scheduling request
- OpenAI: ~$0.05 per request
- Groq: ~$0.005 per request
Ollama is completely free.
Q: Can I use multiple providers?
A: Yes! Configure fallback chains:
ai:
fallback:
enabled: true
providers: [ollama, claude, openai]
AI will try Ollama first, fall back to Claude if needed.
Q: Does AI work offline?
A: Yes, if you use Ollama (local provider). Cloud providers (Claude, OpenAI, Groq) require internet.
Q: How accurate is AI scheduling?
A: AI suggestions are validated through:
- Timezone conversion checks
- DST transition validation
- Working hours verification
- Conflict detection
All suggestions include confidence scores.
Q: How does pattern learning work?
A: Pattern learning analyzes your historical calendar events to identify:
- Meeting acceptance patterns (which days/times you prefer)
- Duration patterns (how long meetings actually run)
- Timezone preferences (which zones you collaborate with most)
- Productivity patterns (best times for focus work)
Requirements:
- Minimum 10-15 historical events for basic patterns
- 30+ days recommended for reliable insights
- Works with completed meetings only (past events)
Usage:
# Analyze last 30 days
nylas calendar ai analyze --days 30
# Quarterly review
nylas calendar ai analyze --days 90
Privacy: All analysis happens locally. LLM calls only for recommendations (opt-in).
Q: Why am I getting "no events found" for pattern learning?
A: Pattern learning analyzes past events only. If you see "no events found":
- Check date range: The default is last 30 days. Your calendar may only have future events.
- Try longer period:
nylas calendar ai analyze --days 60 - Verify calendar access: Ensure the CLI can access your calendars
Note: Pattern learning needs historical data. It cannot analyze future events.
Q: Does pattern learning analyze all my events?
A: Pattern learning fetches up to 200 events per calendar (Nylas API v3 limit). For most users, this covers 1-3 months of history.
If you have many calendars:
- Analysis includes events from all accessible calendars
- Each calendar contributes up to 200 events
- Total analyzed = 200 × number of calendars
Example: If you have 3 calendars, pattern learning can analyze up to 600 total events.
Tip: For very active calendars (500+ events/month), the analysis focuses on the most recent 200 events per calendar within the specified date range.
Q: Can I delete my AI data?
A: Yes! Run:
nylas ai clear-data
This deletes all learned patterns and preferences.
Q: Is AI required for timezone features?
A: No! All timezone features work without AI:
# Timezone conversion (no AI)
nylas timezone convert --from PST --to IST
# Calendar with timezone display (no AI)
nylas calendar events list --timezone America/Los_Angeles --show-tz
AI adds natural language and predictive features.