Skip to main content

AI DevOps

Legato

Open-source runtime for computer-use agents

  • Community votes0
  • DR by Ahrefs0 no change since last check
  • Started2026

Ahrefs Domain Rating set in type 17 Nov 2026, reset 13 Nov 2026.

An open-source computer-use harness that runs agents 1.48x faster than Anthropic's reference implementation by waiting for the screen instead of sleeping

The story so far

Screenshot of the Legato homepage
Homepage of Legato

Legato is an applied research lab building the runtime between language models and machines for computer-use agents. The open-source harness works with any OpenAI-compatible model like Claude, Qwen, or DeepSeek, and uses a ready-wait mechanism instead of fixed sleeps to achieve 1.48x faster performance than Anthropic's reference loop on the same model.

The platform sends screenshots to a vision-language model, runs the actions the model chooses on a desktop, and decides when the screen is ready for the next screenshot. It batches actions, bounds every call, and works with full desktops including office apps, file managers, terminals, and browsers.

Legato offers multiple components: the open-source core harness (Apache-2.0), legato mcp for cloud desktops compatible with Claude Code and Cursor, legato rl for graded desktop rollouts for computer-use reinforcement learning, and Legato Cloud (beta) with pause, resume, and fork capabilities for production agents and RL at scale.

Typeset features

  • Works with any OpenAI-compatible model
  • Ready-wait mechanism instead of fixed sleeps
  • Batches actions and bounds model calls
  • Desktop pause, resume and fork with memory
  • MCP server for Claude Code and Cursor
  • Graded rollouts for computer-use RL
  • Works with full desktops
  • Supports multiple sandbox providers

Pricing

Pricing model: Open source

Follow in print. Log in to follow Legato in print

Want to see what Legato does next? Follow them for milestones, new products and stories by email.

Legato metrics

Set in type for legato.computer: Ahrefs Domain Rating and Majestic link data, with the change since the last edition.
MetricValue
Domain Rating0 no change since last check
Trust Flow0 no change since last check
Citation Flow0 no change since last check
Referring domains0 no change since last check
External backlinks0 no change since last check

Ahrefs Domain Rating set in type 17 Nov 2026, reset 13 Nov 2026. Majestic set 13 Nov 2026, reset 13 Nov 2026. Link data typeset by Majestic. Trust Flow, Citation Flow and Topical Trust Flow are trademarks of Majestic-12 Ltd.

Legato domain colophon

Read the colophon
From the registry's public RDAP record, checked 17 Nov 2026, 15:02 UTC.
Domain age16 days
Registered
Expires
RegistrarNameCheap, Inc.
DNSSECNot signed

Full colophon for legato.computer

Legato launch colophon

Measured after Legato went live on Zivaro Launch on 13 November 2026.

Domain Rating (Ahrefs)
0 no change since launch
0 at launch (13 Nov 2026), 0 on 17 Nov 2026
Trust Flow (Majestic)
0
First measured 13 Nov 2026
Referring domains (Majestic)
0
First measured 13 Nov 2026

Day 4 since launch. Next receipt: day 30, 13 Dec 2026. These are measurements, not proof of cause: many things move these numbers. Figures we have not measured are left out.

Biggest gains on Zivaro Launch

Your launch, step by step

Free tips from Zivaro Launch for getting ready, launch day, the week after and what to do next.

Read the launch playbook

Questions and answers

What is Legato?

Legato is an open-source (Apache-2.0) computer-use harness that sends screenshots to a vision-language model, runs the actions the model chooses on a desktop, and decides when the screen is ready for the next screenshot. It works with any OpenAI-compatible model.

How is it different from Anthropic's computer-use demo?

Legato runs any model, not only Claude. It waits until the screen is ready instead of sleeping 2 seconds before every screenshot. It sends one screenshot per batch of actions and bounds every call. On the same Claude model it finished office tasks 1.48x faster at the same success rate.

Is it faster than OSWorld's Qwen agent?

About the same speed on fresh tasks, with 19 points more success over two runs (74% vs 55%).

Does Legato include a model?

No. Users can bring any model: Qwen through OpenRouter, Claude, or their own vLLM server.

Can I use it for RL training?

Yes. legato rl collect runs graded rollouts with token logprobs, GRPO groups and per-batch profiles against your own model server.

Leave a review for Legato

No review of Legato set in type yet. Set the first.

Your rating

Reviews are free to set. None are paid for. Read the proofs

Readers' notes · Reprints · Share the page · Your company in print? Claim it · Report a misprint

More on this profile

Legato has a free listing. The owner can unlock the full profile with a paid launch for $5:

  • Founder story (locked)
  • Milestones (locked)
  • Demo video (locked)
  • Dofollow website link (locked)

More pages like this

  1. Number 1: Kimchi Coding logoKimchi Coding

    The governed AI coding platform

    AI DevOpsUnited States

    DR 40 −1 (down) since last check 1 votes. Log in to vote for Kimchi Coding in print
  2. Number 2: Spooky Particles logoSpooky Particles

    Monitors Vercel deployments and sends failures to your coding agent for automatic fixes

    AI DevOps

  3. Number 3: AI TrustBOM logoAI TrustBOM

    Git-native AI procurement evidence for your software

    AI DevOpsUnited Kingdom

    DR 0 no change since last check 0 votes. Log in to vote for AI TrustBOM in print
  4. Number 4: Precipiq

    The AI Consequences Ledger

    AI DevOpsUnited States

    DR 0 no change since last check 0 votes. Log in to vote for Precipiq in print
Help centre