Roadmap

We would rather tell you where things stand than let you find out by clicking. This page is the single place that says what exists, what is being built, what is being designed, and what is still only a direction.

Available now

You can use all of this today. None of it needs an account, and none of it is a promise.

WhatWhere
The tnt command — a single static binary, no runtime to installCLI reference
The platform API contract — base URL, resources, error shapePlatform API
The vocabulary these docs useGlossary
The security model — isolation, data handling, abuse and copyrightSecurity
This documentationDocumentation

The CLI and the API are the parts you can build tooling against today. The API itself is being deployed alongside the sandbox, so treat the paths as provisional while the label on each page says so.

In development

This is the first thing we are building, and it is what the platform is for.

WhatWhere
The sandbox — an isolated machine on our compute, with SSH accessThe sandbox
tnt box create / list / exec / destroyThe sandbox
tnt box ssh-proxy ssh — a local proxy that carries an ordinary SSH session to the sandboxThe sandbox
Sandbox lifecycle — idle reclamation, snapshots, destructionThe sandbox
A walkthrough, from an empty laptop to a running agentGetting started

The point of the sandbox is the thing it fixes: an agent that stops when your laptop sleeps. You get a machine, you get a shell, you run your own commands, and it keeps running whether or not you are connected.

Being designed

We are working on the design now. It is a large piece of work and we are not going to pretend it is further along than it is.

WhatWhere
The agents platform — agents you define, running for real on sandboxesCloud agents
tnt agent create / run / logs / stopCloud agents
Agent-to-sandbox attachment, schedules, and access controlCloud agents

On the roadmap

These are directions, not designs. They are further out, and they change as we learn what the first two pieces actually need.

WhatWhere
Shared memory — a memory layer your agents share, organised around an ontologyShared memory
Further tools, chosen by what the sandbox and the agents platform make obvious—
Specification-driven delivery — the model drafts a specification, you approve it, the platform builds from the approved snapshot—

Why this order

It is worth explaining, because the order is unusual.

The tempting way to build this is to start with the part that looks like the product: the agents platform, with a dashboard and an agent that does something visible. That is the demo. It is also the part that depends on everything else working.

We started at the bottom instead. An isolated machine that runs your work, that stays up, that you can reach over SSH, and that is destroyed when you are finished — that is the hard part. Everything above it needs the same substrate: an execution environment you can trust, reproduce, and throw away.

The agents platform is a client of that substrate, not a replacement for it. When it ships, an agent will run inside a sandbox. You will still be able to get a shell and look at what happened — which is the same access you have today.

Starting at the bottom also means the thing we ship first is useful on its own. A remote machine for an agent that has to keep working is a real answer to a real problem, and it does not need anything above it to be worth anything.

How to read the labels

Every page in this documentation carries one of four states, unless it is a normal page with nothing to declare:

LabelWhat it means
(nothing)Available now.
In developmentBeing built. Not available yet.
Design documentDescribes what we are building. Not a feature that exists today.
On the roadmapA direction we have committed to exploring. Not designed in detail, and not built.

If a command appears in these docs and it does not run, that is a bug in the docs, not in your setup — tell us and we will fix it.

Changes

We record substantive changes to the platform and these terms here, and we announce material changes to the terms by email as well. The last change to this page was the reorganization of the toolkit into the four stages above.

Next: the documentation index.