Fray.run

I want to...

Choose your path — we'll show you what matters most.

~/developer

For Developers

Orchestrate agents

Multiple agents in real Kubernetes pods, sharing a live workspace and coordinating with each other — so orchestration never becomes your job. BYO keys, repos, and environments.

~/creator

For Creators

Build with AI agents

Speak your idea and watch it come alive in your browser — and understand it as it grows. Real software, no code required.

Built From a Workflow We Live In Every Day

The infrastructure is Kubernetes-native. What you feel is a team that coordinates itself.

⚙

Multi-Agent Orchestration

Run Claude Code, Codex, Pi, and Shell agents at once, each in its own Kubernetes pod. They hand off to each other through a flexible coordination primitive — discovering each other, dividing work, checking details — so you set direction, not routing. Fray sits above the harnesses you already trust.

↔

Configurable Workspace Sharing

Agents in a workspace share one real filesystem and collaborate in real time — and yes, they can step on each other, on purpose, the way a real team does. Or keep them isolated for independent work.

◉

Flexible Agent Networking

Per-project network isolation today, plus an encrypted per-workspace mesh for secure access from your own devices. Arbitrary agent topologies — pipelines, meshes, isolated pods — are on the way.

⊞

BYO Everything

Bring your own LLM keys, your git repos, and your project's real dev environment — detected from your Tiltfile, Docker Compose, or Kubernetes config.

☁

Deploy Anywhere

Kubernetes-native and portable by design — no single provider can turn you off. When you're ready to ship, Fray wires CI/CD to your platform or an in-house path through self-hosted GitLab.

Architecture Topologies

shared-workspace topology

Claude Code

Shell Agent

↕ ReadWriteMany PVC ↕

Shared Codebase Volume

All agents read and write to the same volume in real-time

handoff topology

Planner Agent

→

Coder Agent

→

Test / Review Agent

Sequential hand-off between specialized agents

isolated topology

Agent A
own volume

Agent B
own volume

Agent C
own volume

✕ no shared state ✕

Fully isolated pods — separate volumes, no shared state

The Hard Part Was Never the Code

🧩

Real projects have real moving parts

A serious idea is rarely one file. It's services, front ends, data, and infrastructure that all have to work together — and most AI tools stop at a snippet.

🧭

Coordination is the tax

Juggling repos, environments, and sessions is the work that isn't the work. It's what makes building feel slow.

🔌

You shouldn't have to be an operator

Clusters, dependencies, dev servers — running the environment pulls you away from the thing you're actually making.

AI Agents That Actually Build Things

Multiple AI assistants working together to turn your ideas into real, working software.

Describe what you want

Tell the AI what you're trying to build — by voice or text. A landing page, an app, an automation — whatever you need.

They divide the work between themselves

One plans, another builds, another tests — handing off to each other directly, without you routing every step.

Get working software

Not code snippets — actual working apps you preview live in your browser as they take shape.

What Can AI Agents Build?

🚀

Ship a product

From MVP to a multi-service suite — accounts, databases, billing — running in front of you.

💼

Take on real client work

Deliver production software, not prototypes.

🧱

Model a whole system

Multiple repos, front ends, and back ends as one coordinated body of work.

⚡

Automate the busywork

Wire your systems together and let the manual steps disappear.

You're Always in Control

👁

Working or waiting — you always know

You always know whether an agent is working or waiting, and you can watch it live. Because your history lives in Fray, not the chat app, it's auditable — with tool-call and log history as we extend it.

✋

Guardrails, not handcuffs

Isolated workspaces let agents move fast. Set guardrails structurally — through hooks and roles — or use our defaults. And one tap stops any run.

💬

Speak naturally, learn as you build

As you collaborate with agents, you'll build understanding of your system alongside the system itself.

🔍

Review it together, as stakes rise

Fray will help you review what was built — grounded in source and ADRs, surfacing what deserves a closer look. Understanding, not just output. On the roadmap.