The Multiplayer Coding Agent Workspace

HumanLayer is a multiplayer coding agent IDE and cloud that works seamlessly from your workstation to anywhere you want to run background agents.

It brings agent sessions, plan artifacts, and code diffs together in one place for your team to collaborate and ship.

HumanLayer ships composable pieces to meet software factory builders where they already are. Bring your own compute, development environment, and agent harness, or run the HumanLayer stack on your workstation or remote workers.

Download⌘+EnterLinux/Windows install instructions

$brew trust humanlayer/humanlayer
$brew install humanlayer/humanlayer/humanlayer

Trusted by engineers at

Multiplayer Coding Agent IDE + Cloud

Bring agent sessions, plan artifacts, and code diffs together so your team can collaborate and ship.

The Platform

Everything you need to ship

A complete AI coding environment — from tasks and artifacts to cloud-scale agent orchestration.

ENG-1234 · Auth refactor

● RUNNINGDesign discussion

DONEResearch auth patterns

IDLEStructure outline

DRAFTImplement JWT

Tasks Group your Sessions and Artifacts

Tasks group agent sessions, artifacts, and worktrees in one workspace that you can collaborate on with your whole team.

QRDSPI · task #1234

✓Q

✓R

▮ DESIGNsynthesizing inputs…

Cutting-edge workflows

The team that brought you context engineering brings frontier AI coding techniques and makes them shareable and collaborative. Battle-tested for solving large problems in million-LOC enterprise codebases.

Artifacts

ARTIFACTS

SUMMARY

●research.md

●design.md

●structure.md

●mockup.html

●plan.md

Every artifact in one place

Research, designs, plans, mockups. Versioned, shareable, and tied to the task.

design.md

## Summary

## Approach

★

Comment-driven refinement

Humans and Agents collaborating on design docs. Catch mistakes before 2,000 lines of code.

session · running

✓Editauth.ts

function refresh(token) {

return fetch(URL, {...})

return mutex.run(() => {

return fetch(URL, {...})

})

}

Full session visibility

You can’t do advanced context engineering without staying in touch with your context. HumanLayer optimizes for high-level orchestration, while letting you drill into every thinking message, every subagent tool call, and every code change.

real-time topology

☁ cloud daemon

📱 mobile

HumanLayer API

💻 coworker

💻 your workstation

local daemon

●

From your laptop to the cloud

A local daemon runs many Agent sessions in parallel on your laptop. Spin up cloud daemons for long-running work. Everything syncs through the HumanLayer API — same UI on web, desktop, or phone, with your teammates one tab away. Agents keep working even if you close your laptop.

workspace

›jump to session_

New task

task⌘N

Jump to session

nav⌘K

Open artifact

doc⌘⇧A

Send to agent

action⌘⏎

Comment on selection

collab⌘/

Superhuman for claude code

Jump between tasks, sessions, and artifacts with a keystroke. A command palette for every action.

Multiplayer Coding Agent IDE + Cloud

Tasks/ENG-1234 Auth middleware refactor

Tasks

Auth middleware refactor

API rate limiting

User dashboard redesign

Database migration v2

Search relevance tuning

Sentry alert routing

Feature flag cleanup

Task NameSHARED WITH ORGENG-1234

Auth middleware refactor — token refresh improvements

/Users/dev/projects/platform-api

Status

Session

Updated

DONE

Research authentication patternsResearch

10m ago

●RUNNING

Design discussion: token refresh strategyDesign

2m ago

IDLE

Structure outline: middleware moduleStructure

15m ago

IDLE

Implementation plan: JWT validatorPlan

20m ago

DRAFT

Implement JWT validation middlewareImplementation

1m ago

ARTIFACTS

SUMMARY

AUTO

●research.md

●design-discussion.md

●structure-outline.md

●mockup.html

●implementation-plan.md

]next tab

TASK

01 / 08

Tasks Group your Sessions and Artifacts

Tasks group agent sessions, artifacts, and worktrees in one workspace that you can collaborate on with your whole team.

↓ Scroll to explore

Watch the Demo

Do not outsource the thinking

Ensure alignment at every step and put engineers in the driver's seat for code quality and architecture.

Structured workflows give the agent every opportunity to show you what it's wrong about before moving to implementation.

QRSPI

A workflow that earns trust

Six phases that turn fuzzy intent into shipped code, with checkpoints your team can verify.

Questions

Generate clarifying questions before any code is written.

Research

Map the codebase, dependencies, and surrounding patterns.

Design

Draft a design doc your team can comment on, in real-time.

Structure

Break the design into phased, verifiable implementation steps.

Plan

Detailed plan with file paths, test cases, and acceptance criteria.

Implement

Execute with continuous feedback. No 2,000-line surprises.

Do not outsource the thinking.
Every phase is a place to push back.

✓Questions

✓Research

▮Design

Structure

Plan

Implement

Technical Design: Human + Agent Collaboration

Gathering codebase context and research findings...

Codebase

src/

index.ts

app.tsx

lib/

utils.ts

api.ts

components/

Button.tsx

Agent

▮ Design Doc

## Summary

## Approach

You

Developer

Product Manager

Tech Lead

Agent

↓ Scroll to see the workflow

Comment-Driven Design Reviews

Inline comments on design documents. Your whole team — humans and Agents — collaborating in real-time. Catch mistakes before they become 2,000 lines of code.

Why not just use a Notion or Google Doc? Because there, your design doc is disconnected from the work. In HumanLayer, comments and decisions feed straight back to the agents — the doc is the interface to the code, not a separate artifact you hope someone reads before shipping.

Design Review

Catch mistakes before code

Your whole team — humans and Agents — commenting on the design before a single line is written.

DESIGNsidebar.md

## Summary

## Approach

Kyle✓ resolved

This separation is causing the S/shift+S hotkey conflict. Can we share the same pure component?

Agent✓ resolved

Good catch — updated the design to use a shared SidebarWidget with configurable tabs.

Sarah✓ resolved

Can we move the sidebar to the left? Matches our other pages, puts navigation closer to the cursor.

Alex✓ resolved

We need to call the API for spline reticulation — widget can’t compute that locally.

Agent✓ resolved

Right — adding a SplineAPI call between setActiveTab and persist.

Alex✓ resolved

+1, this matches the session detail pattern. LGTM to proceed!

Design Review0/7 resolved

Team opens the design document...

DESIGNsidebar-consolidation.md

Summary

Consolidate task and session sidebars into a shared component.

Task view should use the same multi-tab sidebar pattern as session view.

Persist per-task UI state (open, selected artifact, active tab) to localStorage.

Current State

- TaskDetailSidebarPanel: artifact viewer only, no tabs

- ConnectedSidebarWidget: multi-tab (minimap, artifacts, todos, summary, tips)

- S/shift+S cycles TaskDetailWidget tabs (NOT sidebar) — causes hotkey conflict

- Opening artifact replaces sidebar entirely — tabs disappear

- Mini artifacts table takes up horizontal space alongside TaskSummary

UI Mockup

riptide.app/tasks/core-1211

IN PROGRESS

Artifacts

Summary

Auto

layout: sidebar-right

Technical Design

Desired End State

- Shared SidebarWidget with tabs: Artifacts, Summary, Auto-Advance

- Artifact viewer nested within artifacts tab, with back button

- Fullscreen artifact mode takes over entire sidebar

- TaskDetailWidget removed — top section becomes full width

- Sidebar docked to the right side of the task view

What we are not doing

- Todos tab (no single conversation context at task level)

- Minimap tab (no conversation events to visualize)

- Tips tab (RPI phase derived from session.labels, not task)

- New zustand store — keep useLocalStorage pattern

Comments

↓ Scroll to see the review

Spin Up a Task in Seconds

Describe the work, pick a workflow, and let HumanLayer provision worktrees and launch the first agent session.

Flexible task creation and worktree management for multi-repo workspaces.

Tasks/New Task

Start with a Task

Describe what you want to build inline, or pull it straight from a Linear or Jira ticket. HumanLayer drafts the task name and slug for you as you type.

↓ Scroll to explore

From the team that brought you Context Engineering

Leading experts on getting the most out of AI coding agents.

No Vibes Allowed

GitHub YouTube

Solving hard problems in complex codebases. AI Engineer Code Summit, November 2025.

RPI to QRSPI

Blog YouTube

What we learned from thousands of engineers using our open-source workflows.

12 Factor Agents

GitHub YouTube

The original "context engineering" repo, April 2025.

INVESTORS FROM

Pricing

Simple plans for individuals, teams, and enterprises.

Starter

Free for small teams

Everything in Pro, for small teams:

✓ Up to 3 team members
✓ Up to 200 sessions / month

Pro

$100/user/mo

Includes:

✓ BYOK for Claude Code, Codex, and more
✓ Unlimited tasks and sessions
✓Multi-Repo Workspaces & Worktrees
✓Inline Diagrams & Mockups
✓ Real-Time Human+Agent Collaboration on Artifacts
✓Compounding Engineering — Agents get smarter over time
✓ Remote Daemons and other Software Factory building blocks

Enterprise

Custom

Everything in Pro, plus:

✓ SSO/SAML
✓ Audit Logs
✓ Volume Pricing
✓ Custom Terms
✓On-Prem & Private VPC

FAQ

Do I bring my own AI subscription, or pay for tokens?

Bring your own. Plug in your existing Claude, Codex, and other AI subscriptions or API keys (BYOK) — there's no separate HumanLayer per-token bill to reason about.

Which agents and models are supported?

Claude Code, Codex, Copilot, Fireworks, with new models/subscription providers added regularly.

Why not just use Notion or Google Docs for design reviews?

You can, but those are part of a fragmented, manual, pre-AI world. In HumanLayer, comments and decisions feed directly back to the agents, so your design docs live in the IDE, directly integrated with the implementation.

I already have my own skills and workflows

Today, HumanLayer supports a variety of RPI and experimental workflows, and we're actively working on ways for you to bring your own skills and workflows into HumanLayer.

Why not just stitch these workflows and tools together myself?

You can — but you'll spend months in the exploration phase trying things and seeing what sticks. HumanLayer is an opinionated system of agents that you can stand up in 10 minutes. We build HumanLayer with HumanLayer, so we know it works, and we've spent hundreds of hours working in the trenches with customers making the prompts and workflows "squishy" - that is, flexible to a broad range of problems and inputs, so there's no holding it wrong.

$brew trust humanlayer/humanlayer
$brew install humanlayer/humanlayer/humanlayer