Your Google Workspace, unified. Powered by an AI agent that can actually act.
Drive · Gmail · Calendar · Tasks — in one proactive dashboard.
|
|
| Smart mail workspace | Task board |
|
|
| Calendar timeline | |
FlowSpace is an open-source personal dashboard for Google Workspace. It surfaces everything that needs your attention — emails, meetings, overdue tasks, flagged items — in one operational view, paired with an AI agent that can actually do things.
It runs as a local web app with zero cloud infrastructure. Your data stays between your machine and Google's APIs. Note that prompts sent to the AI provider (OpenAI, Anthropic, etc.) are processed by that provider — choose one you trust.
- Proactive dashboard — Briefing summary, active flags, next meeting, and source health at a glance. Clicking flags scrolls directly to flagged items.
- AI daily briefing — Morning summary with attention items, meeting prep notes, reply priorities, and deadline alerts.
- AI chat agent with 23 tools — Streamed responses with structured result blocks:
- Read: search Drive, read Gmail threads, fetch calendar events, list tasks, read Sheets ranges
- Write (approval-gated): send email, create calendar events, write to Docs, append to Sheets, upload files
- Workflow: standup report, meeting prep, weekly digest, email-to-task
- Smart inbox triage — AI-categorized emails: needs reply, needs input, FYI, can ignore. Per-email actions: draft reply, accept/reject meetings, create tasks.
- Follow-up tracker — Tracks commitments across Gmail and Calendar. Snooze, complete, or delete.
- Configurable AI provider — OpenAI, Anthropic, OpenRouter, LM Studio, or any OpenAI-compatible endpoint. Set in the Settings UI.
- "Reply to Sarah's email about Q4 budget — say we're on track"
- "Find the project proposal doc from last week"
- "Schedule a 30-min sync with the team next Tuesday afternoon"
- "Summarize my unread threads and flag anything that needs a reply today"
- "Create a task: follow up with legal by Friday"
Write operations always show an approval step before anything is sent or saved.
Early beta: FlowSpace is actively evolving. Google sign-in works out of the box for all users — no setup required.
- Node.js 20+ — nodejs.org
- A Google account (personal or Workspace)
- An AI API key — OpenAI, Anthropic, OpenRouter, or a local model (optional for read-only use)
- Platform: Tested on macOS. Linux should work. Windows requires WSL.
curl -fsSL https://raw.githubusercontent.com/mohamedelrefaiy/flowspace/main/install.sh | bashThen run in the terminal:
flowspaceFlowSpace opens at http://localhost:3000 and walks you through first-time setup.
Click "Sign in with Google" in the app. FlowSpace will:
- Install the
gwsCLI if it isn't present - Open your browser for Google OAuth consent
- Import credentials — you're in
No GCP project, no client secrets, no environment variables needed for auth.
Go to Settings → AI Provider, choose your provider, and paste your API key:
| Provider | Where to get a key |
|---|---|
| OpenAI | platform.openai.com |
| Anthropic | console.anthropic.com |
| OpenRouter | openrouter.ai |
| LM Studio | Local — no key needed |
Settings are saved to ~/.flowspace/.llm-settings.json. You can change providers at any time.
To update to the latest version, re-run the installer:
curl -fsSL https://raw.githubusercontent.com/mohamedelrefaiy/flowspace/main/install.sh | bashTo uninstall:
rm -rf ~/.flowspace && sudo rm /usr/local/bin/flowspacegit clone https://github.com/mohamedelrefaiy/flowspace.git
cd flowspace
make install
make dev| Task | Command |
|---|---|
| Dev server (HMR) | make dev |
| Production build | make build |
| Production server | make prod |
| Run tests | npm test |
| Type check | make typecheck |
| Docker build + run | make docker && make docker-run |
┌──────────────────────────────────────────────────────────────────────┐
│ YOUR MACHINE (localhost:3000) │
│ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Browser — React 19 · Tailwind CSS v4 · Framer Motion │ │
│ │ │ │
│ │ Dashboard · Inbox · Calendar · Tasks · Chat · Automations │ │
│ └──────────────────────────────┬───────────────────────────────────┘ │
│ │ HTTP / SSE │
│ ┌──────────────────────────────▼───────────────────────────────────┐ │
│ │ Express Server (Node.js) │ │
│ │ │ │
│ │ ┌─────────────────────┐ ┌──────────────────────────────────┐ │ │
│ │ │ AI Agent │ │ Workflow Scheduler │ │ │
│ │ │ │ │ │ │ │
│ │ │ tool-call loop │ │ setInterval per workflow │ │ │
│ │ │ (up to 5 rounds) │ │ email triggers · auto-triage │ │ │
│ │ │ │ │ failure tracking + retry │ │ │
│ │ │ 23 tools │ └──────────────────────────────────┘ │ │
│ │ │ approval gating │ │ │
│ │ │ memory retrieval │ ┌──────────────────────────────────┐ │ │
│ │ └──────────┬──────────┘ │ Persistent Storage (local) │ │ │
│ │ │ │ │ │ │
│ │ │ │ ~/.flowspace/ │ │ │
│ │ │ │ ├─ memory (embeddings) │ │ │
│ │ │ │ ├─ conversation summaries │ │ │
│ │ │ │ ├─ workflow state │ │ │
│ │ │ │ └─ credentials (gws) │ │ │
│ │ │ └──────────────────────────────────┘ │ │
│ └─────────────┼────────────────────────────────────────────────────┘ │
└───────────────-┼──────────────────────────────────────────────────────┘
│
┌────────┴─────────┐
│ │
▼ ▼
┌───────────────┐ ┌───────────────────────────┐
│ Google APIs │ │ AI Provider (your key) │
│ │ │ │
│ Gmail │ │ OpenAI · Anthropic │
│ Drive │ │ OpenRouter · LM Studio │
│ Calendar │ │ any OpenAI-compatible │
│ Tasks │ └───────────────────────────┘
│ Sheets │
└───────────────┘
Everything runs locally — no cloud infra, no data leaving your machine except to Google's APIs and your chosen AI provider. A single Express process serves the API and the React frontend. The AI agent and workflow scheduler share the same server process; all state is written to ~/.flowspace/ on disk.
Most "AI + productivity" tools are thin wrappers: you type, the LLM responds, nothing actually happens. FlowSpace is built differently — the LLM is a reasoning core wrapped by systems for safe execution, persistent memory, and autonomous scheduling.
When you send a message, the agent doesn't just reply — it reasons in rounds. Each round it can call tools, inspect results, and decide what to do next. This repeats up to 5 times before returning a final answer.
User: "Find the Q4 budget doc and summarize the key numbers"
Round 1 → search_drive("Q4 budget") → returns file list
Round 2 → read_document(fileId) → returns doc content
Round 3 → (no more tools needed) → generates summary
This means the agent can handle multi-step work — search, read, cross-reference, then act — without you having to break it into pieces.
Read tools (search, fetch, list) run instantly. Write tools (send email, create event, edit doc, append to sheet) always pause and show you an approval card first — with editable fields, so you can adjust before confirming. Nothing is sent or saved until you explicitly approve.
Agent: "Here's the draft reply to Sarah. Edit if needed — approve to send."
[Draft body] [Edit] [Approve] [Cancel]
The UX is not "are you sure?" — it's "here's what I built, make it yours."
Beyond one-shot commands, FlowSpace supports multi-step workflows defined as sequences of tool calls with output chaining. Each step can reference the output of any previous step:
Workflow: "Log credit card alerts to spreadsheet"
Step 1: gmail_search("from:bank subject:alert") → threadIds
Step 2: gmail_read({{steps.0.threadIds}}) → email bodies
Step 3: sheets_append({{steps.1.parsed_amounts}}) → [approval required]
The scheduler runs these workflows on a configurable interval — autonomously, in the background — and tracks failures with retry state so nothing silently drops.
The agent remembers context across sessions. When you reference a file, a contact, or a decision — even from weeks ago — it retrieves the relevant memory using semantic search (embedding similarity, not keyword matching), so you don't have to re-explain your context every time.
Long conversations are automatically summarized when they approach the token limit, preserving key decisions and open questions without truncating history.
Everything runs on your machine. The server process, the Google API calls, the memory store — none of it touches a cloud service you don't control. The only external calls are to Google's APIs (for your data) and your chosen AI provider (for reasoning). Swap providers anytime from the Settings UI.
Contributions are welcome. The codebase is TypeScript throughout with a clean separation between the Express API, the AI agent, and the React frontend.
Good places to start:
- Add a new AI tool in
src/agent/tools.ts— follow the existing pattern (define schema, implement handler, add to tool map) - Improve triage heuristics in
src/lib/triage.ts - Add a new dashboard panel component under
src/components/
Before submitting a PR:
make typecheck # tsc --noEmit must pass
npm test # all tests must passOpen an issue first for anything beyond a small fix so we can align before you invest the time.
MIT — see LICENSE.
If FlowSpace saves you time, a ⭐ on GitHub goes a long way.