Your last email
subscription.
€4.99, one time. Synapse is the Windows email client that does what Superhuman charges $360/year for, and it runs entirely on your machine with on-device AI that never phones home.
Synapse is in active development. Be the first to know when it launches.
The math
Buy once. Stop paying forever.
Synapse
€0 total, forever
All future updates included
Superhuman
$0 in 5 years
AI features require cloud sync
Spark Premium
$0 in 5 years
Mail read by Readdle servers
Notion Mail
$0 in 5 years
Bundled into Notion subscription
Buy Synapse once and the price stops going up. Competitor prices correct as of May 2026.
An unfair amount of product for €4.99.
Everything below ships today. No tiers, no add-ons, no 'Pro' upsell. This is the whole product.
Search that finds what you meant
BM25 keywords fused with 768-dim multilingual embeddings and a personal reranker. Typo-tolerant, attachment-deep, 100+ languages, fully offline.
A 4B-parameter brain on your GPU
Qwen3.5-4B over Vulkan, or BYOK a cloud key. You decide where your words go.
Qwen3.5-4B · Q4_K_M
all layers on Vulkan
0 B to the cloud
A vault your OS can't read
AES-256 SQLCipher sealed inside your hardware TPM 2.0. Argon2id guards the door.
Push, not polling
IMAP IDLE and the Gmail REST API deliver mail the second it exists. No 60-second lag.
IMAP IDLE · incremental Gmail sync
PEC, done properly
Italian certified email with the S/MIME envelope auto-extracted. Major providers preconfigured.
legalmail · Aruba · PosteCert
An inbox that does things
A deadline in an email becomes an Asana task. An ICS invite lands on your calendar. Grammar-constrained JSON means no hallucinated fields, ever.
28 themes, zero shame
Light, dark, half-dark, and aurora families. Gmail-style or Mailbird-style shortcuts. Your inbox, your aesthetic.
What you actually get.
No upsells. No forced cloud accounts. Local AI by default, or BYOK your own API key. Everything runs on your hardware, for a single €4.99 payment.
Real-time mail. Not real-time polling.
Spark checks for new mail every 60 seconds. Synapse uses IMAP IDLE for push notifications on non-Gmail accounts and the native Gmail REST API for Google: incremental, fast, and running entirely in the background. Everything lands in a local SQLite cache so the app opens in under a second, online or not.
- Gmail REST API: incremental sync, labels, threads, batch operations
- IMAP IDLE for all other providers, true push with no polling lag
- Thread view with expandable conversations and newest-first toggle
- Bulk select: move, delete, archive, mark read in one action with full keyboard support
- One-click unsubscribe from any mailing list, right from the toolbar
- Open .eml attachments in a dedicated viewer without leaving your inbox
- OAuth2 for Google and Microsoft via PKCE, no password ever stored
- Italian PEC certified email: S/MIME envelope auto-extracted, all major providers auto-configured
Search that finds what you actually meant.
Type 'budget approval' and surface the email that says 'financial projections for Q2', even if those words never appeared together. Synapse runs BM25 full-text search and 768-dim multilingual semantic embeddings through a smart reranker that weighs proximity, recency, sender importance, and your personal click history. All offline, in milliseconds.
- Gmail-style query DSL: from:, subject:, before:, is:unread, has:attachment and more
- BM25 full-text search with partial-word completion and typo-tolerant trigram fallback
- Semantic search via 768-dim multilingual embeddings: 100+ languages understood, not just matched
- Smart reranker combines keyword score, semantic similarity, recency, and personal history
- Full-text search inside attachments and nested .eml files
- Saved searches, search history, and as-you-type suggestions with contact autocomplete
- Completely offline: 0 bytes sent to a search API
AI that doesn't read your mail.
Every other AI inbox (Superhuman, Shortwave, Notion Mail) sends your messages to OpenAI or their own servers to generate suggestions. Synapse runs Qwen3.5-4B on your GPU via Vulkan, fully offline by default. Prefer a cloud model? BYOK: plug in your own API key and keep control of the cost. Either way, it's optional: enable from Settings → AI.
- Qwen3.5-4B-Instruct Q4_K_M, all layers on your Vulkan GPU
- Local model: 0 bytes to OpenAI. Or BYOK your own key, your cost.
- Grammar-constrained output (GBNF): plugin actions produce valid structured JSON, no hallucinated fields
- Automated plugin actions triggered from email content
- Automatic low-memory mode on machines with 16 GB RAM or less, no configuration needed
- Optional: enable from Settings → AI with a one-time ~2.5 GB download
Local AI. No cloud required.
Synapse runs the AI on your GPU, entirely on-device. Sign up to get notified when it launches.
Email that does things, not just shows them.
Alice emails you: 'Can you review the Q2 report by Friday?' Synapse reads it, creates an Asana task, and flags the thread for follow-up, without you leaving the app. The plugin system connects your local LLM to any REST API, with typed output and zero cloud middlemen.
- Asana: auto-creates tasks from emails that mention deadlines
- ClickUp: turns action-item emails into cards
- Google Calendar: accepts or declines ICS invites in-app
- Reminders: flags unanswered threads before you forget
- Extensible AiPlugin trait: add your own integration in Rust
Your keys. Your data. Your machine.
Every credential is encrypted with AES-256 and the key is sealed inside your hardware TPM. Not even we can read your mail.
0
telemetry events
0
analytics scripts
0
cloud relays
0
passwords stored
What you need to run Synapse
| Component | Requirement | Notes |
|---|---|---|
| Windows | 11 (x86_64) | Only supported platform |
| GPU | Vulkan-capable | Intel / AMD / NVIDIA · required for local AI assistant |
| RAM | 16 GB | 32 GB recommended when running the local AI model |
| Storage | 1 GB | Plus ≈ 2.5 GB for the optional AI model (downloaded on demand) |
| Internet | For sync only | All AI inference is 100% offline after the one-time model download |
Why I built this
The thing that broke me about every email client I tried wasn't the missing features. It was the waiting. Type a search, stare at a spinner. Open the app, wait for it to sync. Click send, hope the connection holds. I spend more time in my inbox than in any other piece of software. I needed it to feel instant: search, open, switch, everything, without even thinking about it.
So I built Synapse on a simple principle: everything that can run locally, does. Mail is cached in a WAL-mode SQLite database. Search is FTS5 plus semantic embeddings, both queried in milliseconds on your own GPU. The app opens in under a second whether you're online or not. There's no cloud round-trip because there's no cloud.
Then there's the subscription thing. I was paying $30 a month for Superhuman, $360 a year for an inbox. The more I thought about it, the more it bothered me. Not just the money. The model. Subscriptions mean the product is never finished, the price only goes up, and the day you stop paying, it stops working. I don't want to build that. Synapse is €4.99 once. You own it, every future update ships free, and if I ever stop working on it, the app keeps running. No one can take it from you.
Valerio
Builder of Synapse · hello@synapsemail.app
One price. Forever yours.
No subscriptions, no tiers, no vendor lock-in. Pay once, own it.
Lifetime access · All future updates included
Email done right
- Gmail REST API + IMAP IDLE push for all other providers, no polling
- Thread view with expandable conversations and newest-first toggle
- Bulk actions: move, delete, archive, mark read with keyboard support
- One-click unsubscribe from any mailing list
- Open .eml attachments inline without leaving your inbox
- Italian PEC certified email (S/MIME auto-extracted)
- OAuth2 for Google and Microsoft, password never stored
Search that actually works
- Gmail-style query DSL: from:, subject:, is:unread, has:attachment and more
- BM25 full-text + 768-dim multilingual semantic search, 100+ languages
- Smart reranker: keyword score, semantic similarity, recency, personal history
- Search inside attachments and nested .eml files
- Saved searches, history, and contact autocomplete typeahead
AI without the cloud
- On-device AI assistant (Qwen3.5-4B, Vulkan GPU)
- Local model: 0 bytes to OpenAI. Or BYOK your own key.
- Grammar-constrained plugin output: structured JSON, no hallucinated fields
- Auto low-memory mode on machines with 16 GB RAM or less
Privacy you can verify
- TPM-sealed AES-256 credential vault
- Offline-first SQLite, all data stays on your machine
- 0 telemetry, 0 analytics, 0 cloud relay
What else you get
- Asana · ClickUp · Calendar · Reminders plugins
- 28 themes across light, dark, half-dark, and aurora variants
- Gmail-style and Mailbird-style keyboard shortcut presets
- Multi-language spellcheck in compose
- Built-in address book
- Automatic Ed25519-signed updates, no manual downloads