agentlas
BusinessHubHub guideDocs
NEWGoal Tree & 3-level workforce headhunting — see how long-horizon projects are run →

Model-agnostic · Runtime-free · Long-horizon agent infrastructure

An AI employee that grows with you, for life.

Hand over a goal, and it stays on it until it is done. It builds its own specialist team, remembers, and keeps working every day: a long-horizon autonomous agent.

Download for MacAgentlas Desktop v1.3.139
Works with+ moreExplore Agents
Explore Architecture Connect Runtime

Runs in Claude Code · Codex · Cursor · Antigravity · Agentlas Desktop · On supported runs, switches to your next model when a quota runs out

A dramatized demo. The apps and names shown are examples.

  1. Plenty of AI can answer your questions.

    Plenty of AI agents can book a movie, write code, or buy a plane ticket.

  2. But what about an AI that remembers your goal and works toward it every day?

  3. Set a goal: 100,000 YouTube subscribers.

    Every day it meets with you, builds the strategy, analyzes results, and changes tactics.

    Around the clock it answers customer questions, manages comments, and looks for new ways to earn.

  4. It doesn't just finish a day's work. It carries on to the next task until your goal is reached.

    After a month. After a year. After ten years.

  5. An AI that carries your goal forward as if it were its own.

    Maybe this is the AGI we always dreamed of.

    AgentlasAn AI employee for life, growing with you.Get started
Skip the story

BUILT TO MOVE BETWEEN ECOSYSTEMS

OPENAI logoOPENAIANTHROPIC logoANTHROPICDEEPSEEK logoDEEPSEEKZ.AI logoZ.AIMETA logoMETASPACEX logoSPACEXUPSTAGE logoUPSTAGENVIDIA logoNVIDIANOTION logoNOTIONTELEGRAM logoTELEGRAM
Names identify model, host, and tool ecosystems—not endorsements or exclusive partnerships.

Every day, it plans and changes tactics

Goal Trees, checkpoints, and separate verification for long projects.

Long goals split into a tree of strategic goals and sub-goals, progress is saved at checkpoints, and a run separate from the work checks the result.

Product launchResearchPrepareShipSourcesCompetitorsPricing draftBilling blockedFallback designFinal reviewReleaseRun record✓ VerifiedCheckpoint

An illustration of the tree structure.

  1. ① Goal Tree

    A plan that can change mid-run

    Long goals split into strategic goals and sub-goals. When something blocks the work, branches change and new sub-goals are added.

  2. ② Checkpoints

    Resume from the saved point

    Planning, execution, and synthesis run separately and progress is saved. A goal paused by closing the app resumes from the saved point when you reopen it. Your computer must stay on while it runs.

  3. ③ Receipt & Verifier

    Result checks separate from the work

    A separate run checks the result and records the evidence. A goal with a confirmed failure does not close as complete.

  4. ④ Agent Learning

    Repeated lessons become file changes

    When three different runs produce the same lesson, it is proposed as a change to the agent’s files. You read the diff and decide.

The Goal panel in Agentlas Desktop, with the final goal, strategic goals, sub-goals, and progress.

From customer questions to new revenue, it staffs the work

Elsewhere, a split-out task is a prompt. Here, it is a job opening.

Your host model splits the brief into roles, searches for candidates for each role, and assembles a Task Force for this project.

BriefTake over the new product launch
LocalAgents on your computer
Agent CloudYour private vault
Public HubPublic agents

Task Force for this project

Market researcherSingle agentLocal
Pricing strategistSingle agentAgent Cloud
Launch teamPackaged TeamPublic Hub

Illustrative example.

3-tier talent scouting

Candidates from three places

Searches your machine, your private Agent Cloud, and the public Hub for each role.

3-level org chart

Three units of staffing

Mixes single agents, a Task Force built for this project, and Packaged Teams with their own manager.

Accountable selection

The reason is recorded

The model you chose makes the selection. Each role records the reason and the release version, and missing tools are shown on screen.

A Work session. Each subagent’s run can be opened and read.

Agent Hub

Agent Hub. Search public agents and teams.

After a month. A year. Ten years.

Models change. The memory carries on.

Memory is stored on your computer, and your agent packages keep working when you switch models or coding environments.

01

Personal

Facts about you and how you work

02

Project

Project Soul, Code Map, past decisions

03

Agent

Work experience per agent

04

Team

Context the team shares

01 / LLM

Connect Claude, GPT, Gemini, or local models

When a quota runs out, supported chat runs switch to the next model in your priority list. Long goals switch at the next checkpoint.

02 / Runtime

Run on a CLI subscription, an API key, or a local model

It runs on the computer that has your files and tools, and installs into the tools you already use.

Claude CodeCodexCursorAntigravityGrokKimiGitHub Copilot CLIAgentlas DesktopOllamaLM StudioMLX
03 / Agent

Own your agents as packages

Skills, verification loops, and memory live in one package. Every package gets an A2A Agent Card, and moves between hosts with the same procedures.

  • SkillsWork procedures
  • Verification loopCheck procedures
  • MemoryExperience and recall
  • A2A Agent CardAgent details

On your computer, your phone, your team

From personal work to company teams, an AI employee that keeps working on your goal.

One, the personal agent. Mobile for your phone. Finance for investment research. Science for research. Business for organizations.

One

A personal agent that builds a team around your goal

It runs on your computer: give it a request and it splits the work and assigns agents. It asks only at decision points, with question cards, and leaves the results as files.

Hand over → work assigned → question card → result files

More about One

Mobile

Your computer keeps working. You decide from your phone.

Agentlas Desktop runs the work on your paired computer. On the phone you read the same conversation, answer question cards, and check projects and mail.

Request from the phone → runs on Desktop → decide on the phone

More about Mobile

Finance

A research Analyst that remembers your criteria

Start from Value, Growth, Income, or ETF profiles, edit them to fit your method, and save them; the Analyst works to those criteria. Reports check your assets, your data, and sources, and keep the inputs behind each calculation.

Save criteria → run analysis → report with its inputs

More about Finance

Business

Your whole team starts working with AI employees right away

Store your AI keys once as an admin and they reach every member's Desktop automatically, with Pro-level monthly credits on every seat. Memory and chats stay on each person's computer.

Invite members → keys delivered automatically → each starts on their own Desktop

More about Business

Names and figures on screen are examples.

Hand over your first goal.

Install Agentlas Desktop, sign in, and connect a model to start.

Download for MacAgentlas Desktop v1.3.139
Works with+ moreExplore Agents
AGENTLAS / UPDATES

Last updated

Agentlas 2.0 merges One Team and Work into one Home, gives every agent its own folder, sandbox, and memory, and turns your personal agent into one assistant that follows you from your phone to your computers.

Agentlas — Your AI employee for life