Applied AI research
The frontier, applied.
We take what AI keeps making possible and turn it into tools for agents.
Why the Labs exists.
Nobody knows how to work with agents yet. Not us, not the labs training the models, nobody. The frontier moves every month, and the manuals get written afterwards.
So we put ourselves where the learning is fastest. Everything at Antasphere is run by agents: our products, our client work, the site you are reading. We spend our days at the exact point where a person and their agents meet, and we feel, earlier than most, precisely where it rubs.
Each time something is missing at that point, we build it. When a tool survives our own daily use, we ship it to yours.
That is the whole method. The Labs is not a roadmap, it is a position: stand at the frontier, work there for real, and turn what holds into instruments any agent can pick up.
Some of what we find out there feels bigger than a product. When it does, we write it down, and it goes in the journal.
The Antasphere team
The instruments.
However capable the models get, they keep reaching for tools, the way they already hand arithmetic to a calculator. Three so far, each one born from a wall we hit in our own work.
- The flagship
Exos
Project management, rebuilt for agents. Everything a project receives becomes one record agents can navigate; every task becomes something an agent can take and ship. Runs every project our Studio delivers.
- clave.work
Clave
The agent IDE. Every session is a terminal; Clave gives the terminals a home: grouped by project, launched in one click, live pages beside the work. Bring whichever agent you already trust.
- slideless.ai
Slideless
Documents agents can ship. Your agent writes the deck or the page; Slideless hosts it, versions it, and hands you the link. A colleague can remix it and send theirs back.
Why the Labs exists
A market is being born: tools for agents.
01
The agents are commoditised.
A loop with access to tools, owned by Anthropic, OpenAI and Google. Nothing left to invent there.
02
The standard for reaching tools exists.
The Model Context Protocol, introduced in November 2024, adopted by OpenAI, Google and Microsoft within a year, handed to the Linux Foundation in December 2025.
03
So the innovation moves into the tools.
Agents will always reach for tools. What they reach for is where the value now goes.
The size of it, in four figures.
-
97M
monthly downloads of the MCP SDKs
March 2026
-
10,000+
public MCP servers in production
Anthropic, December 2025
-
40%
of enterprise applications embedding task-specific agents by end of 2026, from under 5% in 2025
Gartner, August 2025
-
$202bn
of agentic AI spending in 2026, up 141% in a year
Gartner, January 2026
That is the Labs' market, and we are early in it. Three tools shipped, proven every day in the Studio's production, open source as the distribution. The first customers are our own agents.
The Platform
One account, the whole fleet.
Every tool we ship runs on the same platform. Register once and each tool is already yours: one subscription, one identity, nothing left for your agent to negotiate.
- Open your account
Account
One workspace for the whole fleet: your identity, your team and your billing, shared by every tool we ship.
- Read the docs
Docs
Guides and reference for the whole platform, written for people and read daily by agents.
Field notes
Your agent belongs to you.
The intelligence layer is a subscription you already pay, and it answers to you. Products that bottle their own AI inside are betting against that. We take the other side: we equip your agent instead.
Compromise software is ending.
Owning the exact tool your work deserves has become cheaper than renting almost the right one. We think most software ends up written close to its owner, by their agents, and kept.
The artifact is the history.
Interfaces became cheap to regenerate. A dashboard can be saved as it was, data included, and the trail of saved pages becomes the project's history. We think this is a new way to build software, and we are writing it up properly.
When a note grows into something worth reading, it lands in the journal. Read the journal