Synapse — AI-Powered Personal Workspace

Synapse is a modular AI agent built with Streamlit that brings together six intelligent modules into a single personal dashboard. It connects to your Google Calendar, Gmail, and the live web — all running locally on your machine.

---

Modules

🤖 Chat

Conversational AI powered by LLaMA 3.3-70B (via Groq). Supports real-time web search via Tavily, document Q&A via your uploaded PDFs, and calendar/task management — all through natural language.

✅ Tasks

A persistent to-do list. Add, view, and delete tasks directly or through the Chat module.

📅 Calendar

Full Google Calendar integration. Create, update, delete, and recover events using plain English — "schedule a meeting tomorrow at 3pm", "move my Monday standup to 5pm". All times displayed in IST.

📨 Gmail Automation

Automatically fetches unread Gmail messages every 10 minutes via an n8n workflow. For each email:

• Generates a 2-3 sentence AI summary (displayed as a card on the dashboard)
• Generates a professional draft reply, saved directly below the original email in Gmail
• Provides a direct link to open that specific email in Gmail Powered by a local FastAPI backend (synapse_core/server.py) running on port 9000, with Groq LLM summarization, rate-limit retry logic, deduplication by Gmail message ID, and persistent JSON storage.

🔭 Scout

Daily tech briefing auto-generated every morning at 07:00 IST. Searches the web for:

• Latest AI & tech news
• Upcoming hackathons for students
• Technology scholarships
• New developer tools and releases Results are displayed as cards on the dashboard. Can also be triggered manually. The dashboard auto-refreshes every 5 minutes to pick up the latest briefing without a page reload.

📄 Document Analyzer

Upload one or more PDFs and ask questions about them. Uses semantic search (ChromaDB + HuggingFace embeddings) to find relevant passages, then answers using the Groq LLM. Supports multiple documents simultaneously with source attribution.

---

Architecture

Snowy_Synapse/
├── snowy_app.py              # Main Streamlit app (all 6 modules)
├── synapse_core/
│   ├── server.py             # FastAPI Gmail backend (port 9000)
│   └── scout_service.py      # Scout web search + briefing generation
├── data/                     # Runtime data (gitignored)
│   ├── daily_briefing.json   # Scout's latest briefing
│   └── gmail_summaries.json  # Processed Gmail records
├── .env                      # API keys (gitignored)
├── credentials.json          # Google OAuth client (gitignored)
└── token.json                # Google OAuth token (gitignored)

Three processes run simultaneously:

1. python -m streamlit run snowy_app.py — the dashboard (default port 8501)
2. python -m uvicorn synapse_core.server:app --host 127.0.0.1 --port 9000 --reload — Gmail backend
3. n8n start — workflow engine (port 5678) that polls Gmail and feeds the backend

---

Setup

Prerequisites

• Python 3.11+
• Node.js (for n8n)
• A Google Cloud project with Calendar API and Gmail API enabled

1. Clone and install

git clone https://github.com/mrunalC27/Synapse.git
cd Synapse
python -m venv venv
venv\Scripts\Activate.ps1        # Windows
pip install -r requirements.txt

2. Environment variables

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key
TAVILY_API_KEY=your_tavily_api_key

Get your keys from:

• Groq: https://console.groq.com
• Tavily: https://app.tavily.com

3. Google OAuth setup

1. Go to Google Cloud Console
2. Create a project, enable Google Calendar API and Gmail API
3. Create an OAuth 2.0 Client ID (Desktop app type) — download as credentials.json and place in project root
4. Add your Gmail address as a test user under OAuth consent screen
5. On first run, a browser window will open for Google sign-in — this generates token.json automatically

4. n8n Gmail workflow setup

Install n8n globally:

npm install -g n8n

Create a second OAuth Client ID (Web application type) in Google Cloud Console with redirect URI:

http://localhost:5678/rest/oauth2-credential/callback

In n8n (localhost:5678), build this workflow:

Schedule Trigger (every 10 min)
  → Gmail: Get Many Messages (filter: is:unread, limit: 5-10)
  → Wait (2 seconds)
  → HTTP Request POST http://127.0.0.1:9000/gmail
      body: { message_id, thread_id, subject, sender, body (snippet), link }
  → IF: status == "processed" AND draft is not empty
      → Gmail: Create a Draft
          Subject: {{ original subject }}
          Message: {{ $json.draft }}
          Thread ID: {{ original threadId }}
          To Email: {{ original From }}

Activate the workflow — it will run automatically every 10 minutes while n8n is running.

Import the workflow directly

Instead of building the workflow manually, import the pre-built workflow:

1. In n8n, click "Add workflow" → "Import from file"
2. Select n8n/gmail_workflow.json from this repo
3. Open the workflow, click the Gmail credential fields and reconnect your own Gmail account (credentials are not exported with the workflow for security reasons)
4. Update the HTTP Request node URL if needed (http://127.0.0.1:9000/gmail)
5. Activate the workflow

5. Run everything

Open three terminal windows (all with venv activated):

Terminal 1 — Gmail backend:

python -m uvicorn synapse_core.server:app --host 127.0.0.1 --port 9000 --reload

Terminal 2 — Dashboard:

python -m streamlit run snowy_app.py

Terminal 3 — n8n:

n8n start

Open http://localhost:8501 in your browser.

---

API Keys needed

Key	Purpose	Free tier
Groq	LLM for chat, summarization, draft replies	Yes — 30 RPM on llama-3.1-8b-instant
Tavily	Web search for Scout and Chat	Yes — no preset limit
Google OAuth	Calendar + Gmail access	Yes

---

Known limitations

• All three processes (Streamlit, uvicorn, n8n) must be running simultaneously for full functionality. No persistent background service is set up by default — closing any terminal stops that component.
• Groq's free tier caps at 30 requests per minute. The Gmail workflow uses a Wait node and retry logic to stay within limits, but processing very large backlogs (50+ unread emails) in one run may hit rate limits temporarily.
• Scout's 7 AM auto-briefing only fires if Streamlit is actively running at that time (uses APScheduler inside the Streamlit process). For guaranteed daily runs regardless of whether the app is open, set up a Windows Task Scheduler job pointing to python -c "from synapse_core.scout_service import run_scout; run_scout()".
• The Document Analyzer's ChromaDB vector store (chroma_db/) is local only and not committed to git — re-upload your PDFs after a fresh clone.

---

Tech stack

Layer	Technology
Frontend/UI	Streamlit
LLM	LLaMA 3.3-70B, LLaMA 3.1-8B via Groq
Web search	Tavily
Gmail automation	n8n + FastAPI
Vector search	ChromaDB + HuggingFace all-MiniLM-L6-v2
Google APIs	Google Calendar API, Gmail API
Scheduler	APScheduler (in-process)
Backend	FastAPI + Uvicorn

---