Teams maintain infrastructure, not product
Shipping a service means CI/CD, servers, storage, backups, monitoring, security, scaling, and licensing — dedicated engineers and ongoing cost that slows the product down.
Early access · French Tech
An IT department in your pocket.
Whatever you do, describe the task — Tabbify builds, deploys, and runs it on a distributed cloud. Teams ship in one command. Everyone else just asks.
- Ship to production in one command
- Or just describe it in plain words
- Bring your own LLM — ChatGPT, Claude, Mistral
- Data stays in your country
- Free tier, no credit card
hugo.tabbify.app asking
H
I run a barbershop in Wissembourg. I need a booking site with WhatsApp and Telegram confirmations.
T
I'll assemble it from ready services, host it on compute near you, and ask before any customer message goes out.
T
Approve domain, WhatsApp, and Telegram for this workspace?
Hugo's booking flow · domain, WhatsApp, Telegramneeds approval
domain:connectwhatsapp:sendtelegram:send
H
Approve. Use my logo and keep Sundays closed.
The problem
Software stopped being the hard part. Running it didn't.
For almost any business task the program already exists. The difficulty is everything around it — getting it running, wiring it together, and keeping it alive in the right place. Five gaps, the same root.
Ready software you still can't just use
Mature open source exists for almost any task, but launching it, wiring local payment and tax systems, and maintaining it for years takes a whole team. The gap is run, connect, maintain.
Compute is expensive and far away
GPU for AI and ML is costly, concentrated among a few large providers, and often nowhere near the client's region. Getting the right resource quickly and cheaply is hard.
Small business needs one tool, has no IT
An in-house hire sits idle and costs too much; a contractor turns every change into weeks of requests. A small business needs a single specific tool, not a department.
Infrastructure that doesn't fit the region
Big clouds and PaaS are either absent from the region or fall under foreign law (the US CLOUD Act) — at odds with data-localization rules already in force in many countries.
Two audiences, one platform
For people who code. And people who never will.
The platform is built for any level of technical skill — the same compute and services underneath, met where each person already works.
Remove the infrastructure routine
A developer writes code and ships it to production with a single command — no manual servers, CI/CD, or monitoring. Focus on the product, not the plumbing.
- Build, deploy, host, monitor and scale in one command
- Connect to ready services instead of deploying from scratch
- Self-host in your own network and jurisdiction, fully isolated
Describe it, the AI assembles it
No DevOps, no database, no developer. Describe the task in plain words and the AI builds the solution on ready infrastructure. A barber asks for a booking form; he gets one.
- Talk to it in ChatGPT, Claude, or Mistral
- Review the result via a preview link, then publish
- A working product from sign-up to launch in about a day
Operate it your waynatural language point and click for engineers
One platform. Any interface you like.
Technical teams work through a panel and an API. Everyone else talks to the platform in plain language — through the apps they already use every day. Same compute, same services underneath.
AI agent via MCP
Talk to the platform inside ChatGPT, Claude, or Mistral. Tabbify connects as an extension over the open MCP standard, so the assistant already knows your context — no new interface to learn.
Visual panel
A single control panel: add an item, change a price, check revenue, sync several locations. The panel can be created and updated by voice — and seen by your whole team.
API and CLI
Everything the panel does, scriptable. Technical teams drive the full lifecycle from their own tools — build, deploy, secrets, logs — and wire Tabbify into existing pipelines.
Building with words
- Describe what you need, in plain language.
- The agent assembles it from ready services.
- Review the result via a preview link.
- Publish — on a domain, on compute near you.
Ready services
Connect to a running system, not an empty server.
You don't deploy ERPNext yourself — you connect to one already running. Tabbify is the single hub where open-source and commercial services are launched and maintained, with licenses folded into one price.
Sa
SambaPOS
open sourcePoint of sale & registers
Od
Odoo
open sourceERP & accounting
ER
ERPNext
open sourceInventory & operations
Me
Metabase
open sourceAnalytics & dashboards
Ke
Keycloak
open sourceAccess & identity
Ji
Jira
commercial · license includedProject tracking
Local integrations included
The private mesh
A private network at the core.
At the centre of the platform is a closed network built on WireGuard. Every user gets their own workspace inside it, wired to the services they need — messengers, open-source systems, your own apps — with no public ports.
Access rights per service
Unlike an ordinary VPN, the mesh connects services directly and sets permissions for each one. Your AI finds the right tool without ever seeing the secret behind it.
A node joins with one command
Any machine — a data centre, an office desktop, an edge box — joins the network with a single command and becomes part of one computing space.
Data flows peer-to-peer
A central panel coordinates the nodes, but data moves directly between them — not through Tabbify. Your own workspace, your own jurisdiction.
The distributed cloud
Compute placed where it makes sense.
One network combines very different machines — data centres, public clouds, and ordinary computers in a home or office — into a single computing space. Compute runs across two kinds of site.
EDGE BOX
Compact hardware near the user, often on consumer-grade equipment. It runs work that tolerates a single node failing — builds, game streaming, ML inference — giving cheap compute close to the client.
Firecracker micro-VMs · light workloadsData centres & public cloud
AWS, Google Cloud, Azure, and our own racks, where reliability matters. GPU workloads run in the Cloud Hypervisor, which passes a real GPU through into the VM for AI and ML.
Cloud Hypervisor · GPU pass-throughThe placement engine
A workload doesn't pick a server — it states what it needs, and the engine places it where it's most advantageous for the client.
pricelatencyjurisdictionenergy
Storage & reliability
Reliability you choose, not hope for.
Databases separate compute from storage — Postgres writes its change log to a quorum of nodes, while the long-term history lives in distributed object storage we build on Garage, under the name Tabbify S3.
How it's built
ComputePostgres, stateless — scales on its own.
Change logWritten to a quorum of nodes.
Tabbify S3Distributed object storage on Garage.
Pick a level
- Survive a machineA single node goes down — no data lost.
- Survive a rackA rack or a power domain fails — you stay up.
- Survive a cityAn entire city goes offline — your data is elsewhere too.
- Stay in the country“Data never leaves the country” — one of these levels.
Status & roadmap
What works today. Honestly.
No vapourware. Here is the real split between what is running in production and what is still being built.
CapabilityStatusComment
P2P mesh — WireGuard, access rights, NAT traversalWorkingA node joins with a single command
App hosting in Firecracker, supervisorsWorkingDeploy with one command and via GitHub Action
Image registry, auth, git forge, agent dev envWorking—
Production on AWS — coordinator, database, CDNWorking—
AI agent (MCP), panel, API and CLIWorking—
Tabbify S3 — distributed storage on GarageIn testingTest deployment done, in active development
Third-party services catalogPartialService publishing exists, full catalog later
Client databases with reliability levels (Neon-like)Under studyDesign and POC, in development
GPU — consumer, via Cloud HypervisorUnder studyLaunch in 2–5 years
EDGE BOX hardware & placement economicsUnder studyModel depends on the country
Smart workload placement — price, energy, latencyPlannedPlacement currently by simple rules
White-label per jurisdictionPlanned—