Le Space Automation

The signs are changing

Standard software had an economic rationale, not a functional one: custom development was too expensive, and running it for years more expensive still. For two generations, companies therefore struck the same compromise, first with the purchased licence, then with the cloud subscription: the organisation adapts to the software, not the software to the organisation. Renting did not resolve that compromise; it made it permanent.

What is changing now is the cost side of that calculation. With AI agents, what used to bind months is built in days, and maintenance, until now the decisive item against custom software, turns from the exception into routine. That moves the threshold at which a solution of your own pays off, and it moves in one direction: not every cloud tool is replaceable tomorrow, but the set of what sustains itself as an in-house solution grows every month.

Two arguments remain that appear in no cost calculation. With rented software your data sits with third parties, often in a foreign jurisdiction; every data-protection review thereby becomes a review of a supply chain. And the bargaining power lies with the provider: if it raises prices, retires the product or is sold, you negotiate without your software and with your data in its hands.

“Data is safer in the cloud” — an argument with an expiry date

The second argument for rented software was safety: backups, redundancy, an operations team. It was a fair one, against the single laptop as much as against the company's own server in the back room, which businesses of every size handed over because patching, backups and round-the-clock duty were not their business. Two things have weakened that argument since.

First, the data centre is not a safe place, only a different one. On 10 March 2021 OVHcloud's SBG2 data centre in Strasbourg burned down; customers whose backups sat in the same building lost their data, and in 2023 a court awarded them damages. On 1 March 2026 an Amazon Web Services data centre in the United Arab Emirates was struck and caught fire; the provider advised its customers to enact their own disaster recovery plans. The lesson is not that data centres are run badly, but that a single copy in a single place remains a risk, however large the operator.

Second, the cloud concentrates what it is meant to protect. One breach at the provider hits every customer at once, and the provider itself sees the data in the clear. A data centre meets the legal requirements; nothing more is promised, and nothing more is delivered.

Our answer: compute and data, back in-house

The principle fits in one sentence: compute returns in-house, and outside the house only ciphertext exists, regardless of whether it is archived centrally or decentrally. That is the core of local-first peer-to-peer, and it is our programme of work.

The data stays with the company and is processed where it resides: on its own devices. Local does not mean one device but the devices of everyone working on the same matter: the people handling a case, the customer, the tax adviser. They carry the same data and reconcile it directly among themselves, device to device. A relay establishes the connection and, when two devices are not online at the same time, holds the changes encrypted until they are; it cannot read them. And archiving happens continuously, with every change, rather than in a backup window at night: encrypted with a key that only the company holds, preferably on decentralised storage, Aleph Cloud today, Filecoin in the works, and on request as a second copy with a classic cloud store such as Amazon S3, Google or Azure. That is the data centre's redundancy without its clear text: a new device gets everything back with the passkey, and the operator holds a copy it cannot read.

Only computing tasks that cannot be done in-house leave it, and then exclusively into environments whose protection goes beyond the legal requirement. The most important case today is the language model. It runs locally; where that is not enough, with a provider of your own choosing, hosted centrally or decentrally, and then inside a trusted execution environment (TEE) whose attestation can be verified: the operator can read neither input nor output. For that we currently prefer decentralised compute, from Aleph Cloud and the Akash Network among others.

Said plainly, all of this holds once a second device or the archive is set up; a single computer without an archive is as vulnerable as ever. Which is why archiving is part of the set-up with us, not of the recommendations. You keep browser and operating system current yourself; we patch no servers, because there are none.

What we automate

• Receipt flow and pre-accounting — with Le Space Belege as a ready building block: collect receipts, match payments, hand over the DATEV package
• Quotes and invoicing, with payment matching
• Post and mail intake: sort, extract the data, file, draft replies — never sent unchecked
• Data transfers between systems that do not know each other: CSV, interfaces, portals
• Recurring reports and evaluations
• Replacing single subscription tools with a tailored tool of your own

How the agents work

• Agents write and maintain the code; people set the goals and check every result. We answer for what runs — not the model.
• Every change lives versioned in Git: traceable, revertible, with its reason.
• Automated tests are the net: whatever the agents change has to prove nothing breaks before it reaches you.
• Open source wherever possible — checkable instead of taken on trust. The value is not in the code but in it running, and in someone standing behind it.
• The language model stays yours, not ours: local, or in a verified environment of your choosing, as described above. Le Space runs no AI and sees none of your data.

Step by step, not big bang

We start with one process, not your whole IT. The current solution keeps running alongside until the new one has proven itself in daily use — only then is the next process due. And you can cancel without loss: if you cancel us, you keep the software, the source code and your data. With a subscription you keep an export file.