Multi-agent orchestration platform

Lead armies of AI agents, sanely.

Deploy hierarchical teams of agents with persistent identity, memory, continuity and cross-collaboration tools.

The Taurus app: an agent tree with live statuses, schedules, and run history
01 Introduction

What is Taurus?

Taurus is a powerful multi-agent AI orchestration platform.

Before Taurus, power users of AI were juggling 20 Terminal tabs, with Claude Code, Codex, and so on. Each one stateless, none of them talking to each other.

On Taurus, they are able to seamlessly orchestrate hundreds of agents into trees, projects, mini-companies, run them on schedules, and stay sane while doing all that.

02 Mechanics

How?

Our answer to “What is an agent?”

Before Taurus, people would talk about “a Claude agent”, “a Gemini agent”, “a Codex agent”. On Taurus, you can have:

Orion engineering lead

Your engineering lead for your startup.

├─ implementer — does the legwork
├─ critic — reviews code
└─ visioner — architecture brainstorming
Vega-CEO runs a news company

Runs your automated news company “Vega” on a schedule.

├─ researcher-main · researcher-alt
├─ editor · publisher
└─ email-responder
Aristotle research lead

Leads an agentic tree focused on frontier math research.

Clara personal assistant

Your personal assistant that handles calendar and comms.

These are all real examples from our users.

How do I split them?

The most important difference between agents is context. Orion has its own memory, knowledge base, instructions, container, filesystem, and local repositories that are all about engineering for your company. Orion doesn't need to know what Vega is doing.

But how do they collaborate?

An agentic tree is what makes Taurus different from every other tool out there.

When you create an agent as a child of another, they become part of a tree. The parent can delegate tasks to children, review their work, send them back for revisions, and coordinate across the team. Children can have their own children, going as deep as you need.

All agents in a tree share a /shared folder, like a company Google Drive. When they need to pass files, collaborate on a repository, or leave notes for each other, that's where it happens.

You give the parent a task. It figures out how to break it down, who does what, and whether the results are good enough. You check in when you feel like it.

Real containers

Every agent runs in its own isolated environment with a full Linux shell, a persistent filesystem, and a browser. They install packages, run servers, write files, clone repos. Everything in their workspace persists between runs.

03 Everything else

What else?

Agents running on schedules, waking up at 9am, reading their memory, picking up where they left off. Static sites and dashboards maintained by agents and hosted for you. 30+ models from every major provider, mixed freely across your team. A full REST API for building on top of Taurus. Voice input, file attachments, five editor themes, a web IDE with terminal access.

03.1
Schedules
Cron-based, with overlap handling and memory continuity
0 9 * * * · overlap: skip
03.2
Dashboards
Static sites built and maintained by agents, hosted for you
/shared/public/site → hosted
03.3
30+ Models
Anthropic, OpenAI, Google, xAI, Groq, OpenRouter
claude · gpt · gemini · grok
03.4
REST API
SSE streaming, blocking ask, WebSocket terminals
POST /v1/agents/orion/runs
03.5
Web IDE
Editor, terminal, file browser, five themes
editor · terminal · files
03.6
Remote Targets
Connect agents to your machines via SSH or relay
ssh://build-box · relay

Create your first agent in under a minute

Start building.