outlook-daily-brief

Outlook Daily Brief

OpenHome Ability that acts as a daily orchestrator: when you say a trigger phrase, it pulls your Outlook calendar, unread email, and local weather in parallel, then uses an LLM to turn everything into one natural-sounding ~60-second spoken summary. It’s not for one-off queries like “what’s on my calendar” — it’s the morning briefing that combines all three sources into a single cohesive readout.

Works with Outlook.com (consumer) and Office 365 (enterprise). Uses Microsoft Graph (OAuth) for calendar and mail, and Open-Meteo (free, no key) for weather.

---

What the code does

1. Trigger and intent
The ability is invoked by phrases like “give me my brief”, “good morning”, “start my day”, or “what did I miss”. The code reads self.worker.agent_memory.full_message_history to get the user’s last message(s), then classifies intent:

• Full brief — “good morning”, “brief me”, “daily brief”, “start my day” → standard ~60s morning briefing.
• Urgent / catch-up — “what did I miss”, “catch me up” → same data but a shorter, urgency-focused script (~30–40s) that leads with what needs attention.

2. Parallel fetch
Calendar, email, and weather are requested at the same time (each in a thread via asyncio.to_thread), with a per-call timeout (~5–6s). If one source fails, the others still run; the briefing uses whatever came back.

• Calendar — GET /me/calendarview for today (start/end in UTC). Returns subject, start, end, location. Scope: Calendars.Read.
• Email — GET /me/mailFolders/inbox/messages with isRead eq false, top 5, subject/from/receivedDateTime. Scope: Mail.Read.
• Weather — Open-Meteo forecast by lat/lon. Location comes from saved prefs or auto-detect (e.g. ip-api.com); result is stored so the next run reuses it.

3. LLM synthesis
All fetched data (calendar, email, weather) is passed in one payload to text_to_text_response() with a system prompt. The LLM returns a single script to be read aloud — not three separate readouts. Different prompts are used for full brief vs “what did I miss” so the tone and length match the intent.

4. Speak and follow-up loop
The script is spoken via the capability worker. Then the code waits for the user:

• Repeat — re-fetches data, re-synthesizes, and speaks again (or replays last script if fetch fails).
• Change my city to [city] — updates location in prefs, then “say repeat to hear your brief again”.
• What did I miss / catch me up — re-fetches and speaks the urgent-style brief.
• Stop / done / quit / bye / etc. — says goodbye and calls resume_normal_flow().

Idle timeouts (no input for several cycles) trigger a short “I’m still here…” message and then exit. Every exit path calls resume_normal_flow().

5. Persistence
Preferences are stored in a namespaced file (e.g. outlook_daily_brief_prefs.json): location, calendar_connected, email_connected, enabled_sections (weather, calendar, email). Sections can be turned off; missing or failed sections are skipped in the brief without being announced.

6. Errors
Each section fails on its own. If every fetch fails, the user hears: “I’m having trouble reaching some services right now. Let me try again in a moment.” No crash; the ability resumes normal flow.

---

Setup: Client ID and Refresh Token

Configure CLIENT_ID and REFRESH_TOKEN in main.py (and keep TENANT_ID = "consumers" for personal Microsoft accounts).

Use the refresh_token folder for step-by-step setup:

1. refresh_token/README.md — Create an Entra app, add delegated permissions (Calendars.Read, Mail.Read), enable public client flow.
2. refresh_token/get_refresh_token.py — Set the same CLIENT_ID, run it, sign in at microsoft.com/devicelogin, then copy the printed refresh token into main.py as REFRESH_TOKEN.

Same CLIENT_ID goes in both the script and main.py; the script gives you the REFRESH_TOKEN to paste into main.py.