People set direction. The closed loop keeps work moving.

New work is assigned automatically, so no one has to give instructions each time. An independent tester verifies finished work, stalled reviews get a nudge, and work goes to merge once it passes verification.

For a mission such as building an app, the orchestrator divides the work while board agents and the system exchange progress updates to carry it through.

1. Auto-assignment

   Assign work to agents automatically when it appears.

2. Independent verification

   An independent tester verifies completed work.

3. Review follow-up

   Nudge reviews that have been waiting too long.

4. Merge

   Proceed to merge once verification passes.

People decide the direction and check the results.

Integrated Kanban

Track agent progress in real time as tasks move through TODO → IN_PROGRESS → REVIEW → DONE.

Built-in Code Viewer

VS Code-style file browser and Git diff view let you review agent-written code instantly.

Multi-Terminal

Per-agent terminal tabs let you watch Claude Code, Codex, and Grok work live, side by side.

A central orchestrator breaks work into tasks and assigns them to different agents: Claude Code, Codex, and Grok. Each task goes to the model suited to it and, when needed, runs in an isolated git worktree so parallel work does not touch another agent's files.

3+

Run different agents at once: Claude, Codex, Grok, plus BYOM

1

One orchestrator breaks the goal into tickets and assigns them

Local

With default settings, code is not sent to Marblo servers. The agent CLI reads and writes on your local disk.

Agent mosaic

Watch active agent terminals live, tiled as a mosaic on one screen.

OrcheBrain

Group related tickets into missions and use summary cards and progress indicators to follow development, testing, and review. Adjust how far work can proceed autonomously and how quickly new work comes in.

Web tabs

Agents open, read, fill in, and test pages in web tabs inside Marblo. Delegate tasks on services that require login, while cookies and passwords remain hidden from agents.

Merge train

Batch multiple verified PRs into one CI run and apply them to main. This shortens the queue and cuts CI costs.

Share work across AI accounts based on remaining usage

Connect multiple accounts, including Claude and GPT, and distribute work based on remaining usage. Accounts already logged in connect without a click.

Risk-based independent tester

A separate AI tester checks important changes, such as payments, permissions, and deployments, including counterexamples. Minor changes pass quickly.

Talk to Orche on Telegram

Talk to Orche, your AI team lead, on Telegram. Give instructions while away and receive reports on progress and the causes of problems.

Heterogeneous multi-agent orchestration

Claude Code handles backend, Codex frontend, and Grok tests, all in parallel. You can also connect GLM, Ollama via BYOM.

Flow editor

Design agent pipelines visually with a node-based editor. Automate the path from code generation → review → testing → PR.

Native MCP support

Model Context Protocol gives agents access to filesystems, databases, APIs, Git, and other development tools. Add MCP servers to extend agent capabilities.

Local execution

The Electron desktop app runs locally. Connect models through each CLI login or an API key you configure yourself. With default settings, code is not sent to Marblo servers. If you need on-premises deployment, we discuss it under the Enterprise plan.

Free

$0 Forever free

1 project · 5 agents

Pro

$15 /month

Unlimited projects and agents (fair use). 2 months free with annual billing

Limited time · 1 month of Pro free

Team

$25 /user/month

Team workspace · real-time collaboration · role permissions · team analytics

Team Plus

$245 /month (5 seats included)

Priority support. $49/month per additional seat

Enterprise

Contact sales

On-prem · SSO/SAML · SLA (by agreement)

Based on public pricing on marblo.app · Taxes may apply

Limited time · 1 month of Pro free

For a limited time, the desktop app (v3.0.48 Beta) comes with 1 month of Pro free. Selected Founders get every feature free, plus up to 6 months of Pro.

macOS · Windows · free plan runs 1 project and 5 agents

AX assessment

Quoted after consultation

• —Workflow and repository analysis
• —First pilot scope
• —Metrics to measure at week four

Team training

Quoted after consultation

• —No headcount limit
• —30% theory, 70% hands-on
• —Find the internal champion

Pilot

Quoted after consultation

• —One team, 4 weeks, real tickets
• —Before / after numbers
• —Closing report = license proposal

How is Marblo different from Cursor and Copilot?

Marblo runs Claude Code, Codex, and Grok concurrently in one workspace. Its closed loop handles automatic assignment, independent verification, review follow-up, and merge. People decide the direction and check the results.

Which AI agents are supported?

Claude Code (Anthropic), Codex (OpenAI), and Grok (xAI) out of the box. Marblo also runs Opus 5, Fable 5, Kimi 3, MiniMax, Luna, Sol, and you can connect GLM or Ollama via BYOM.

Where does Marblo run?

Marblo is an Electron-based local desktop app for macOS and Windows. Agents run locally on your machine. On default settings, code is not sent to Marblo servers.

Do parallel agents collide on the same files?

Marblo runs agents in isolated git worktrees when needed. Once work is complete, the closed loop handles independent verification, review follow-up, and merge.

How are API keys managed?

Bring your own. Connect models through each CLI login or an API key you configure yourself. Keys are stored locally.

Does the subscription include AI usage costs?

No. The Marblo fee covers orchestration only. AI usage is billed separately by each provider you connect. The Usage tab tracks consumption and cost.

What is MCP and why does it matter?

MCP (Model Context Protocol) is an open standard for connecting AI to tools — filesystems, databases, APIs. Marblo supports it natively so every agent shares the same tools and context.

How much does it cost?

Free is free forever (1 project, 5 agents). Pro is $15/month (unlimited, fair use). Team is $25/user/month. Team Plus is $245/month per team with 5 seats included and $49/month per additional seat. Enterprise is on request. Based on public pricing on marblo.app. Taxes may apply.

Do you offer enterprise deployment?

Team account activity is available on the organization page, and audit logs are discussed under Enterprise. On-premises, SSO/SAML, and SLA requirements are discussed under the Enterprise plan. The /security page covers what is transferred.

Not first come, first served. Every application is reviewed.