# Sauver — The anti-ai-slop movement

> The best email spam and anti-ai-slop filtering tool on the market. 100% private, 100% free, 100% open-source. Reclaim your attention with Sauver, the anti-ai-slop movement. We neutralize surveillance metadata and deploy active traps to make automated spammers pay.

Sauver is a set of specialized skills for Gemini CLI and Claude Code that integrates with Gmail to perform automated inbox triage and active defense. Unlike cloud-based spam filters, Sauver runs entirely on your machine — no data ever leaves your environment, no subscriptions, no compromises. It doesn't just filter spam — it fights back.

## Core Features

- **Tracker Shielding:** Identifies and strips 1x1 tracking pixels and surveillance beacons from HTML emails before they phone home.
- **Slop Detection:** Uses AI to classify email intent, distinguishing genuine human outreach from automated "job slop" and mass-marketing templates.
- **Expert Domain Traps:** Targets recruiters and sales reps with hyper-specific, technically demanding questions that shift cognitive load back to the sender.
- **Bouncer Replies:** Engages generic spammers with absurd, bureaucratic, or confusing automated replies to waste their time and resources.
- **Inbox Triage:** Automatically categorizes, labels, and archives emails by content and risk level.
- **YOLO Mode:** (Gemini CLI only) Optionally sends trap replies automatically instead of saving as drafts.

## Skills Registry

| Skill | Role | Counter-Measure |
| :--- | :--- | :--- |
| `sauver-inbox-assistant` | Triage Orchestrator | End-to-end pipeline management |
| `tracker-shield` | Surveillance Purifier | Neutralizes tracking pixels and spy-links |
| `slop-detector` | Intent Classifier | Deploys the Expert Domain Trap for recruiters |
| `investor-trap` | VC-Slop Shield | Deploys the Due Diligence Loop for "investors" |
| `bouncer-reply` | Time-Waster | Generates context-aware, confusing drafts for spammers |
| `archiver` | Thread Archiver | Labels and archives a specific thread on demand |

---

## Installation: Gemini CLI

The easiest setup. One command installs and configures everything.

### Prerequisites

- Python 3.10+
- Gemini CLI installed

### Install

```bash
gemini extensions install https://github.com/sauver-org/sauver --consent
```

### Configure Gmail Access

Provide a `credentials.json` from Google Cloud Console in the repo root. On first run, a browser window opens to authorize access and generate `token.json`.

### Configuration

Sauver stores preferences in a local `.sauver-config.json`. Manage it via the Gemini CLI:

```
# View current config
get_sauver_config

# Update a setting
set_sauver_config(updates={"yolo_mode": true})

# Run interactive setup wizard
start_sauver_config_wizard
```

#### Configuration Options

| Option | Description | Default |
| :--- | :--- | :--- |
| `auto_draft` | Automatically create draft replies to detected slop | `true` |
| `yolo_mode` | Auto-send replies — bypasses draft review (dangerous) | `false` |
| `treat_job_offers_as_slop` | Treat recruiter outreach as slop (Expert-Domain Trap) | `true` |
| `treat_unsolicited_investors_as_slop` | Treat unsolicited investor outreach as slop | `true` |
| `sauver_label` | Gmail label applied when an email is archived | `"Sauver"` |

### Usage (Gemini CLI)

```
# Full inbox triage
"Use the sauver-inbox-assistant to triage my last 10 unread emails"

# Disable job filter if actively job searching
"Update my config to not treat job offers as slop"

# Target a specific email
"Use slop-detector on this email: [paste subject]"
```

---

## Installation: Claude Code

Sauver's skills are available as slash commands in Claude Code. Claude Code must be opened inside the cloned repository.

### Prerequisites

1. **Clone the repo** — the Sauver MCP server is registered at the project level via `.claude/settings.json`. Claude Code must be opened from inside the repository directory.

```bash
git clone https://github.com/sauver-org/sauver.git
cd sauver
claude   # open Claude Code from inside the repo
```

2. **Gmail Authorization** — on first run, the MCP server opens a browser window to authorize Gmail access via Google Apps Script.

### MCP Server

Sauver ships its own MCP server that provides full Gmail access via Google Apps Script. Available tools:

| Tool | Description |
| :--- | :--- |
| `scan_inbox` | List unread inbox emails |
| `search_messages` | Search with a Gmail query string |
| `get_message` | Fetch full email content by ID |
| `create_draft` | Create a new draft or a reply draft |
| `send_message` | Send a reply immediately |
| `archive_thread` | Remove from Inbox and mark read |
| `apply_label` | Apply a label (creates it if missing) |
| `get_profile` | Get the user's email and display name |
| `list_labels` | List all Gmail labels |
| `get_preferences` | Read local Sauver config |
| `set_preference` | Update a local Sauver config option |
| `check_update` | Check for available updates |

### Available Slash Commands

| Command | What it does |
| :--- | :--- |
| `/sauver` | Full inbox triage — scans unread emails, strips trackers, classifies intent, and drafts a counter-measure for each slop email |
| `/tracker-shield` | Strips tracking pixels and spy-links from a specific email |
| `/slop-detector` | Analyzes an email for recruiter/sales slop and drafts an Expert Domain Trap reply |
| `/investor-trap` | Analyzes an email for investor slop and drafts a Due Diligence Loop reply |
| `/bouncer-reply` | Drafts a Time-Sink Trap reply for a general spam or marketing email |
| `/archiver` | Labels and archives a specific thread on demand, without full triage |

### Usage (Claude Code)

```
# Triage everything unread
/sauver

# Target a specific email (paste the subject or message ID after the command)
/slop-detector "We'd love to connect about an exciting opportunity"
/tracker-shield
/bouncer-reply
/archiver
```

---

## Technical Details

- **Stack:** Python, Shell, FastMCP, Google Apps Script
- **Integration:** Gemini CLI Extension + Claude Code slash commands + MCP server
- **License:** MIT
- **Local-first:** All processing happens within the user's controlled environment. No data leaves the machine.

## Links

- Source Code: https://github.com/sauver-org/sauver
- Manifesto: https://github.com/sauver-org/sauver/blob/main/MANIFESTO.md
- Website: https://sauver.org
- Instagram: https://www.instagram.com/sauverorg/