Compare · The AI Chief of Staff

Keep your AI.Murage makes it a whole team.

Murage is a free, open-source AI Chief of Staff for the AI you already pay for. Brief it once. It builds a team of bots and runs each one on the engine that suits it (Claude Code, Codex, Hermes or a local model), on your own plans. Change an engine and the team, its memory and its rules stay put. New bots ask before they act.

Download Murage See the side-by-side

Free app under AGPL-3.0. Your AI plan is separate. Runs while your computer is on. macOS, Windows and Ubuntu.

It runs the real Claude Code and Codex.
You sign in to the unmodified Claude Code on Anthropic’s own screen; Murage never reads your login. Codex is the official Codex CLI you install and sign in to yourself. Hermes runs through its own ACP adapter.
How Murage runs Claude Code
Change the engine, keep the team.
A bot’s engine is a setting. Its name, memory, skills, routines and approvals live in Murage, so they stay when you switch.
The stop line is code you can read.
New bots start on Ask. On Full access, a bot stops before messaging anyone new, posting publicly, paying, or deleting outside its folder, unless you have allowed it for that task. Replies in an existing conversation go ahead on Auto and Full access. No limits turns the stop line off. On Hermes and other ACP engines, Ask and the stop line cover only what the engine asks Murage about: shell commands and file edits.
Read server/stop-line.ts

Log off at 11pm. Wake up to a fix to review. A freelance developer in a forest cabin twice: slumped at his laptop at 11pm, then relaxed with coffee at 7am while Ember lists a fix to review, a client reply and an invoice.

Already using one? Keep it. Murage makes it better.

Murage doesn’t replace the agent you pay for. It runs it as a bot’s engine and adds the team around it.

Murage + Claude Code
Runs inside Murage as a bot’s engine
Keep Claude Code. Murage makes it a team of specialists.
What Murage adds
- A Chief of Staff above it
- Every engine on one team
- Built for the rest of the business
Free on top of your Claude planFull comparison
Murage + Codex (ChatGPT Work)
Runs inside Murage as a bot’s engine
Keep Codex. Murage makes it a whole team.
What Murage adds
- A standing team above it
- Memory on your computer
- 1,500+ apps and a stop line
Free on top of your ChatGPT planFull comparison
Murage + Hermes Agent
Runs inside Murage as a bot’s engine
Love Hermes? Keep it. Murage makes it even better.
What Murage adds
- A Chief of Staff and a team
- One window for the whole team
- Ask mode and a stop line by default
Both freeFull comparison

Murage + Claude (Cowork)
Keep using Claude’s app. Murage runs Claude through Claude Code, not Cowork.
Compare
Murage with n8n
Your bots run it as a tool
Compare
Murage with Zapier
Your bots run it as a tool
Compare

Your bots run n8n workflows and Zaps as tools over MCP, and a workflow can start a Murage task through a local webhook (cloud senders need a tunnel). What carries over when you switch

Side by side, gaps included.

Ten questions a buyer asks, answered from each product’s own pages. Where Murage is behind, it says so.

Yes
Partly
No
Not stated

Checked September 26, 2026.

Murage is behind on three of these today: working with your computer off, a native phone app (coming), and a SOC 2 report.

Compare Murage with

What you payBefore usage beyond your plan: Murage:
Free app, plus your AI plan or key
Source
Claude Code:
Paid Claude plan: Pro $20/mo ($17 billed yearly); not on Free
Source
You choose which AI company does the workPer bot or per task, from different vendors: Murage:
Yes
Claude Code, Codex, Hermes, Grok Build, Cursor, Google Antigravity, Kimi, Qwen and local models, per bot. OpenClaw is coming
Source
Claude Code:
No
Claude models
Source
Named agents that keep their memory between sessions: Murage:
Yes
Named bots, each with its own notebook and managed memory
Claude Code:
Partly
Saved subagents with persistent memory folders; agent teams are per session
Source 1 Source 2
Sets up the team from a plain-English requestNo config files: Murage:
Yes
The Chief proposes and builds it
Source
Claude Code:
Partly
Describe teammates and Claude spawns them; experimental, off by default
Source
Asks before acting outside, by defaultMessaging someone new, posting, paying, deleting: Murage:
Yes
Until you choose No limits; replies in an existing thread go ahead on Auto and Full
Source
Claude Code:
Partly
Permission prompts in sessions; cloud Routines run “without stopping for approval”
Source
Memory stored on your computer: Murage:
Yes
Local SQLite, open to read and edit
Claude Code:
Yes
CLAUDE.md, auto memory, subagent memory folders
Source
Keeps working with your computer off, today: Murage:
No
Murage Cloud isn’t open yet
Claude Code:
Yes
Routines on Anthropic’s cloud (research preview, hourly minimum, daily run cap)
Source
Native phone app: Murage:
Coming
iPhone and Android app coming. Today: the web app over Tailscale, plus the Chief in Telegram, Slack and Discord
Claude Code:
Yes
Claude app
Source
SOC 2 report: Murage:
No
No SOC 2 report
Claude Code:
Yes
Anthropic’s, through its Trust Center
Source
Open sourceLicense named: Murage:
Yes
AGPL-3.0
Source
Claude Code:
No
“© Anthropic PBC. All rights reserved.”

Murage compared with Claude Code, Codex (ChatGPT Work), Claude (Cowork), OpenClaw, Paperclip, Grok Bot
What matters	Murage	Claude Code	CodexChatGPT Work	Claude(Cowork)	OpenClaw	Paperclip	Grok Bot
What you payBefore usage beyond your plan	Free app, plus your AI plan or key Source	Paid Claude plan: Pro $20/mo ($17 billed yearly); not on Free Source	Free tier; Go $8, Plus $20, Pro from $100 Source	Paid Claude plan from Pro $20 Source 1 Source 2	$0 (MIT), plus your model costs Source	$0 (MIT), plus the plans of the agents it runs, on a server you run Source	Included with paid Cursor plans (Pro $20) or a SuperGrok link; usage metered weekly Source 1 Source 2
You choose which AI company does the workPer bot or per task, from different vendors	Yes Claude Code, Codex, Hermes, Grok Build, Cursor, Google Antigravity, Kimi, Qwen and local models, per bot. OpenClaw is coming Source	No Claude models Source	Partly OpenAI by default; you can add other providers, incl. local Source	No Claude Source	Yes Dozens of providers, plus Claude Code as a text-only fallback and Codex through its app-server Source 1 Source 2	Yes Claude Code, Codex, Cursor, Bash, HTTP webhooks and OpenClaw Source	Not stated Not stated in its docs Source
Named agents that keep their memory between sessions	Yes Named bots, each with its own notebook and managed memory	Partly Saved subagents with persistent memory folders; agent teams are per session Source 1 Source 2	Partly Custom agents saved as files, each with its own model Source	Not stated Not stated on the product page Source	Partly Multi-agent routing you configure	Partly Agents with roles and reporting lines resume a task’s context; memory is on its roadmap Source	Yes “Names, jobs, and context that compounds over time” Source
Sets up the team from a plain-English requestNo config files	Yes The Chief proposes and builds it Source	Partly Describe teammates and Claude spawns them; experimental, off by default Source	Partly Spawns subagents when asked; custom agents are files you write Source	Not stated Not stated Source	No You configure routing	Partly You set a goal and design the org chart; agents can propose hires you approve Source 1 Source 2	Not stated Not stated Source
Asks before acting outside, by defaultMessaging someone new, posting, paying, deleting	Yes Until you choose No limits; replies in an existing thread go ahead on Auto and Full Source	Partly Permission prompts in sessions; cloud Routines run “without stopping for approval” Source	Partly Approval prompts and plan mode	Partly Deleting needs approval	No Host commands default to full access, ask off; sandbox hosts default to deny Source	Partly Approval workflows and execution policies you set up; per-action defaults not stated Source	Partly “Comes back when something needs your approval” Source
Memory stored on your computer	Yes Local SQLite, open to read and edit	Yes CLAUDE.md, auto memory, subagent memory folders Source	Partly CLI config and custom agents are local files; ChatGPT memory is in OpenAI’s cloud Source	Partly Your files stay in your folders; chat memory in Anthropic’s cloud	Yes Markdown on disk	Yes When you run it locally (embedded PostgreSQL) Source	No Work and context live on its cloud computer Source
Keeps working with your computer off, today	No Murage Cloud isn’t open yet	Yes Routines on Anthropic’s cloud (research preview, hourly minimum, daily run cap) Source	Yes Cloud tasks Source	Yes “Close your laptop, it keeps going” Source	Partly If you run it on a server Source	Partly If you run it on a server Source	Yes “Closing the app, your laptop, or your phone doesn’t stop a background turn” Source
Native phone app	Coming iPhone and Android app coming. Today: the web app over Tailscale, plus the Chief in Telegram, Slack and Discord	Yes Claude app Source	Yes iOS Source	Yes Mobile (beta) Source	Yes Native iOS and Android Source	No Responsive web UI in a phone browser Source	Yes iPhone, iPad and Android Source
SOC 2 report	No No SOC 2 report	Yes Anthropic’s, through its Trust Center Source	Yes OpenAI’s	Yes Anthropic’s	Not stated Not claimed: “no enterprise edition and no paid version” Source	Not stated Not claimed Source	Not stated Not stated Source
Open sourceLicense named	Yes AGPL-3.0 Source	No “© Anthropic PBC. All rights reserved.”	Partly CLI Apache-2.0, app closed Source	No Closed	Yes MIT Source	Yes MIT Source	No Closed

Each mark links the page it comes from. A cell with no link carries our first-party check of September 25, 2026.

Where Murage is behind today

Every limit, with what to do instead

It runs while your computer is on and Murage is open. Murage Cloud isn’t open yet.
No native phone app yet; the iPhone and Android app is coming. Today you reach your Chief of Staff on Telegram, Slack or Discord, and through the web app.
No SOC 2 report today. If procurement needs one, the products marked Yes are ahead.

Weighing an alternative?

Each of these does a similar job a different way, and each is ahead somewhere. Here is who each one suits, in their favor first.

Paperclip
Open source, like Murage
An open-source server for running a company of agents through tickets, with hard spend limits and OpenClaw on the team.
Pick Paperclip if
You want hard spending limits per agent and per goal, today
Pick Murage if
You want to install an app and talk, not run a server
Free (MIT)Full comparison
OpenClaw
Open source, like Murage
A free, open-source personal AI with desktop and phone apps, reachable from 29 chat apps. Coming to Murage as an engine.
Pick OpenClaw if
You live in WhatsApp, iMessage or Signal (29 chat apps; Murage has 3 today, with WhatsApp coming)
Pick Murage if
You want a Chief of Staff to build and run the team, and new bots to ask first (OpenClaw’s host commands default to full access with asking off)
FreeFull comparison
Grok Bot
Hosted, and always on
Named AI teammates from SpaceXAI and the Cursor team that work together on a shared cloud computer.
Pick Grok Bot if
You need your bots working with your laptop and phone closed, today
Pick Murage if
You want Claude Code, Codex, Hermes and Grok Build on one team, on plans you already pay for
From $20 a monthFull comparison
Manus
Hosted, and always on
A general AI agent on a cloud computer that runs tasks, including scheduled ones, on credits.
Pick Manus if
You want to start in a browser with nothing to install, on 300 free credits a day
Pick Murage if
You want named bots that keep a job, and its memory, on your computer, on the AI plan you already pay for
From $20 a monthFull comparison
Genspark
Hosted, and always on
A cloud AI workspace for slides, docs and sheets, plus Claw, an agent on its own cloud computer, on monthly credits.
Pick Genspark if
You want Call For Me to phone businesses for you, and native iOS and Android apps, today
Pick Murage if
You want to run on the AI plan you already pay for instead of monthly credits
Monthly creditsFull comparison

Also compared

What you hand off. What you still decide.

Already getting useful work from Claude or ChatGPT? Keep it. Murage adds a standing team, local memory and one place to review the work.

Clear my inbox
A chatbot: Explains how to write the replies
Murage: Sorts it and drafts every reply in your voice. You press send.
Research a market
A chatbot: Gives you an answer to check yourself
Murage: A team researches and hands you one page with a source next to each claim. Add a reviewer to challenge the numbers.
Follow up with quiet leads
A chatbot: Writes one template
Murage: Drafts a personal note for each lead and ranks who to send first.
Keep an eye on competitors
A chatbot: Can check on a schedule, then leaves the follow-up to you
Murage: A daily routine checks the web and tells you what changed.
Write the weekly update
A chatbot: Summarizes what you paste in
Murage: A Friday routine drafts it in the format you showed it, for you to check.

What you pay, and who you pay.

Murage is free. The AI isn’t. If you already pay for Claude or ChatGPT, adding Murage costs nothing more: your bots use that plan’s allowance, and a busy team uses it faster. If you don’t, you’ll need an AI plan or key, from $20 a month per person for Claude Pro or ChatGPT Plus.

Work out your cost on the pricing page

The hard questions.

Is this allowed under Anthropic’s terms?

Yes. Murage never signs you in, and never sees, stores or passes on your Claude login. It helps you install the official Claude Code, you sign in to Claude Code yourself on Anthropic’s own screen, and Murage runs it as a bot’s engine. Anthropic’s policy says its rules don’t “prevent an end user from signing in to the unmodified Claude Code binary with their own Claude subscription” (Claude Code legal and compliance, checked September 26, 2026).

Your usage counts against your own plan, as it would if you ran Claude Code yourself. A team of bots gets through that allowance faster than one chat, so for heavy work you can give any bot an Anthropic API key instead. Murage is not affiliated with or endorsed by Anthropic.

Isn’t Murage just a wrapper around Claude Code?

Partly, and on purpose. Murage runs the real Claude Code, because Anthropic builds the best agent for Claude. What Murage adds is what one Claude Code session doesn’t have: a standing team of named bots that remember the job tomorrow, Codex and Hermes on the same team, a stop line you can read in the source, your Chief of Staff on Telegram, Slack and Discord, and voice calls. If you only write code in a terminal, Claude Code on its own may be all you need.

Claude Code already has subagents and agent teams.

It does, and they’re excellent inside a coding session. Claude delegates to subagents on its own, and saved subagents can keep their own memory. Agent teams are “experimental and disabled by default”, with “one team per session” (Anthropic’s docs). Murage’s team lives outside any single session: named bots that keep their memory, routines and roles from one day to the next, and that can run on Claude, Codex, Hermes or a local model side by side.

“Free” is the hook. What’s the catch?

The app is free and has no paid tier. The AI isn’t free: your AI provider bills you as usual, and each bot uses your plan’s allowance, so a busy team uses it faster. Murage has no per-bot spend caps yet. Each task can make at most 16 hand-offs per turn and 32 in total. Flux Router is Ferrox Labs’ optional paid service for models, connected apps, voice and web search. Everything it does also works with your own keys. That’s how we plan to make money, along with Murage Cloud when it opens.

Local means it stops when my laptop sleeps.

Yes. Today Murage runs while your computer is on and Murage is open. Every hosted product on this page beats us there. If you need work running around the clock today, pick one of them, or leave Murage running on a desktop that stays on. Murage Cloud, which will run your team on a server we look after, isn’t open yet, and we won’t promise a date.

What happens to my logins? Who are you?

Murage is made by Ferrox Labs, a small team. We have no SOC 2 report or outside security audit yet. What we do have: new bots start on Ask, the stop line is in the source code, logins never travel in a team export, and memory and files are stored on your computer. Stored locally doesn’t mean processed locally: whatever a bot sends to a hosted AI or a connected app is processed there, unless you run a local model. Today’s releases send no usage analytics. Anonymous, opt-out analytics are planned.

AGPL scares my company. Should it?

Running Murage unmodified inside your company doesn’t require you to publish anything, and the license covers Murage’s code, not the work your bots produce. AGPL’s source-sharing duty applies if you change Murage and distribute it, or let other people use your changed version over a network: then you share those changes. It also means nobody, including us, can take the code closed later. This isn’t legal advice; have your counsel read the license.

Is Murage like CrewAI or AutoGen?

No. Those are frameworks developers use to build agents in code, and AutoGen is now in maintenance mode. Murage is an app you download and brief in plain English.

Don’t take our word for it.

Ask your favorite AI about Murage. We’ll open it with the question ready.

“I’m evaluating Murage (murage.ai), a free, open-source desktop app where an AI Chief of Staff runs a team of AI agents on engines like Claude Code and Codex. Using current official sources, what are its weaknesses, who should NOT use it, and how does it compare with OpenClaw, Paperclip, Claude Code and Grok Bot? Separate the app’s price from AI usage.”

Claude, Claude Code, Cowork, Codex, ChatGPT, OpenClaw, Hermes, Paperclip, Manus, Genspark, Grok, SuperGrok, Cursor, n8n, Zapier, Viktor, Lindy, Tasklet and the other product names on this page are trademarks of their respective owners. Murage is not affiliated with or endorsed by them; names and logos are used only to identify the products compared.

Keep the tools you already use. Murage makes them a team.

Give your first job to Murage.

Download the free app, connect the AI you already pay for, and tell your Chief of Staff what needs doing. Plan on about ten minutes from install to a working team.

Free · Download Murage

No account needed · Runs on the AI plan you already pay for

Keep your AI.Murage makes it a whole team.

It runs the real Claude Code and Codex.

Change the engine, keep the team.

The stop line is code you can read.

Already using one? Keep it. Murage makes it better.

Murage + Claude Code

Murage + Codex (ChatGPT Work)

Murage + Hermes Agent

Murage + Claude (Cowork)

Murage with n8n

Murage with Zapier

Side by side, gaps included.

Where Murage is behind today

Weighing an alternative?

Paperclip

OpenClaw

Grok Bot

Manus