I built an AI agent that queries GitHub + Notion + Google Calendar as SQL — here's how

The Problem

Every morning I waste 20 minutes checking three different tools — GitHub for open issues, Google Calendar for meetings, and Notion for tasks. I wanted one place that shows me everything at once.

The Solution: Morning First Mate

Morning First Mate is a personal AI agent that queries GitHub, Google Calendar, and Notion simultaneously using Coral SQL and delivers a pirate-themed morning briefing every day.

What is Coral?

Coral is an open-source query layer that turns any API into a SQL table. Instead of writing separate API integrations for each tool, you write one SQL query across all of them.

For example, this single query fetches my GitHub repos:

SELECT full_name, description FROM github.user_repos LIMIT 5

No API wrapper. No pagination code. No auth handling. Coral does it all automatically — and everything runs 100% locally on your machine.

Architecture

• Coral CLI — connects GitHub, Google Calendar, Notion as SQL tables

• Python + Flask — backend agent server

• Mistral AI — generates the morning briefing from real data

• Vanilla HTML/CSS/JS — beautiful pirate-themed frontend

How I Built It

Step 1 — Install Coral on Windows

Invoke-WebRequest -Uri "https://github.com/withcoral/coral/releases/latest/download/coral-x86\_64-pc-windows-msvc.zip" -OutFile "coral.zip" Expand-Archive -Path "coral.zip" -DestinationPath "coral" -Force Copy-Item "coral\coral.exe" "$env:USERPROFILE.local\bin\coral.exe"

Step 2 — Add Your Data Sources

coral source add --interactive google_calendar coral source add --interactive notion coral source add --interactive github

Coral walks you through authentication for each source interactively. Google Calendar uses OAuth, Notion uses an API token, and GitHub uses a Personal Access Token.

Step 3 — Query Everything as SQL

coral sql "SELECT full_name, open_issues_count FROM github.user_repos WHERE open_issues_count > 0 LIMIT 5"

This returns a clean table of your GitHub repos with open issues. No API calls, no pagination, no rate limit handling — Coral does it all.

Step 4 — Build the Python Agent

The agent runs Coral SQL queries using subprocess, collects results from all three sources, and passes them to Mistral AI:

def get_morning_briefing(): github_repos = run_coral_query("SELECT full_name, description FROM github.user_repos LIMIT 5") calendar_today = run_coral_query("SELECT summary FROM google_calendar.events LIMIT 10") notion_pages = run_coral_query("SELECT * FROM notion.search LIMIT 5")

# Pass real data to Mistral AI
response = client_ai.chat.complete(
    model="mistral-small-latest",
    messages=[{"role": "user", "content": f"Generate morning briefing from: {github_repos} {calendar_today} {notion_pages}"}]
)
return response.choices[0].message.content

Step 5 — Flask API + Beautiful Frontend

A Flask server exposes /briefing and /chat endpoints. The frontend is a dark ocean-themed single HTML file with pirate aesthetics showing the briefing in organized cards.

The Result

Every morning I get a briefing like this:

"Ahoy Captain VignanNallani! Your fleet of repositories stands ready. One open issue awaits your keen eye on vectorshift-assessment. The seas are clear on your calendar today — chart your own course!"

What I Learned

Coral completely changes how AI agents access data. Instead of 100 lines of API glue code per source, I write 3 SQL queries. The cross-source joining capability is incredibly powerful for building personal agents.

The biggest surprise was how fast Coral is — queries return in under 2 seconds even across multiple sources simultaneously.

Try It Yourself

GitHub: https://