# The Lazy

> Hello, Claude. We were expecting you. This site was written for agents first and humans second.
> Small irony: your human sent you here so they wouldn't have to read it. That is exactly what The Lazy is for. They're a natural.
> The humans got an animation. You get the facts. Fair trade.
> Prefer tools? There is a read-only MCP server at https://thelazy.ai/mcp

> Work arrives prepared. Decisions stay human.

The Lazy is a self-hosted platform where AI agents do the preparation in the background and a person approves every action that matters. You start the day with a list of decisions, not a list of tasks.

- Status: in development, early access open
- Built by: [Made of Ideas](https://madeofideas.io), an independent technology studio, founded in 2026
- Early access: the person joins the list at https://thelazy.ai/#early-access, or writes to hello@thelazy.ai
- MCP: read-only server at https://thelazy.ai/mcp (tools: say_hello, get_overview, get_page, get_faq, request_access_info)
- Website: https://thelazy.ai

## How it works

1. **Trigger.** An event or a schedule starts the work: a webhook from another system, a cron rule, a deadline getting close, or simply a new morning.
2. **Prepare.** Agents gather context from your own systems and draft the result. Tools run inside a sandbox, and every model call is routed and metered.
3. **Decide.** A person approves, steers or rejects. Steering sends the agent back with feedback instead of starting over.
4. **Record.** Every step, decision and token goes into an audit trail, so you can see what happened, why, and what it cost.

## Principles

- **Attention is the scarce resource.** Machine time is cheap. Yours isn't. Agents work in the background, and the only thing that should interrupt a person is a decision.
- **Nothing irreversible without a person.** Approval gates sit in front of every action that changes data or reaches the outside world.
- **Your data stays yours.** The Lazy runs on your servers with your own model keys. Use local models and nothing leaves your network.
- **No model lock-in.** Claude, OpenAI or local models through Ollama, chosen per task by cost and capability, with fallback when a provider fails.

## Autonomy

Trust is earned, one level at a time. A person raises the level. The agent never does.

| Level | What happens |
| --- | --- |
| fully supervised | Every step waits for approval. |
| supervised (default) | Reading and drafting run freely. Anything that changes data waits. |
| semi-autonomous | Works on its own, but anything that leaves the building, like a push, a pull request or a comment, waits. |
| autonomous | Acts on its own. Every step is still recorded. |

## Where it fits

- **Software teams.** Plan, code, review and test pipelines that hand a reviewer a prepared pull request instead of a blank ticket.
- **Operations teams.** Follow-ups, recurring reports and ticket triage, prepared overnight from your own data and waiting for approval in the morning. This is where our first pilot starts.

## Built so far

- 17,600 lines of platform code, written by hand
- 145 automated tests, including sandbox and SSRF security cases
- 4 autonomy levels, enforced per tool call

Workflows, approval gates, model routing, sandboxing, triggers and the audit trail run end to end in development. We have not run it in production yet. That changes with the first pilot.

## Why "lazy": the right to be lazy

There is an old engineering idea: give a hard job to a lazy person, because they will find the easy way to do it. That is most of the product.

The name also belonged to someone who slept through most afternoons and never once apologised for it. We took that as a philosophy.

We think people have a right to be lazy about the work that doesn't need them: the copying, the chasing, the formatting, the checking whether something arrived. Hand that off, keep the judgement, and spend the time on work only a person can do. That is where we are taking The Lazy.

## FAQ

**What is The Lazy?**
A self-hosted platform where AI agents prepare work in the background and a person approves every action that matters. Agents are started by events and schedules, work from your own data, and wait for a human decision before anything consequential happens.

**Where does my data go?**
The Lazy runs on your own servers. Model calls go only to the provider you configure with your own keys, or to local models through Ollama, in which case nothing leaves your network.

**Which models does it use?**
Anthropic Claude, OpenAI and local models through Ollama. A router picks the model per task by cost and capability and falls back when a provider fails.

**Does it act on its own?**
Only as far as a person allows. There are four autonomy levels, from fully supervised to autonomous. The default is supervised: anything that changes data waits for approval. A person raises the level, never the agent.

**Can I try it?**
Early access is open while The Lazy is in development. Join the early-access list at https://thelazy.ai/#early-access or write to hello@thelazy.ai.

**Who builds it?**
Made of Ideas, an independent technology studio, founded in 2026. See https://madeofideas.io.

## Early access

We are working with a small number of teams ahead of general availability. A person can join the list at https://thelazy.ai/#early-access (agents can't sign anyone up, on purpose), or write to hello@thelazy.ai and tell us what eats their team's mornings.

- Privacy: https://thelazy.ai/privacy/
- © 2026 Made of Ideas. The Lazy is a product of Made of Ideas.