Most popular
Pro
$29/month
For people running real work every week.
- $15 managed AI usage each month
- 25 hosted workers and 20 active schedules
- 3 concurrent managed runs
- 90-day hosted run history
Portable AI workforce
Build one AI worker.Grow an entire workforce.
Describe the job, connect your tools, and deploy the same portable .agent wherever the work belongs.
Build one portable worker, then grow a workforce across operations, research, growth, and delivery. Every worker is stored as an inspectable .agent file that can open in AgentMug Cloud, Desktop, the CLI, or an MCP host. These are alternative ways to run the same worker.
Your AI workforce
One worker at a time
portable .agent
Company floor4 workers
Built on one runtimeGrows with you
Ready to run
Operations
Morning brief
Finds the mail that needs you and sends one clear briefing.
Starts
Weekdays at 9:00 AM
Moves through
Outlook to WhatsApp
morning-brief.agentyours
Cloud
Desktop
CLI
MCP
Connect the systems your team already uses.
75 executable tools are wired today. Extend your workers with MCP, webhooks, and HTTP when the job needs more.
- Microsoft Outlook
- Slack
- GitHub
- Notion
- Linear
What a worker can actually do
Runs real code in an isolated sandbox
code.execute runs Python in a sandboxed VM with network egress denied — on the managed cloud, never on your machine.
Reads the web with a real browser
web.browse drives a real remote Chromium for pages a plain HTTP fetch cannot read, alongside a vetted Puppeteer screenshot tool.
Full shell on your own machine (AgentMug Desktop)
shell.execute exists only where you run the worker yourself — AgentMug Desktop today, not the CLI. The cloud deliberately refuses it and names a host that can, which is a deliberate split rather than a gap.
75 executable tools are wired today across 20 connected providers. Extend your workers with MCP, webhooks, and HTTP when the job needs more. Read the generated capability matrix
Build it in plain language.
Start with one recurring job. AgentMug turns the description, connections, timing, and review trail into a worker you can keep.
morning-digest.agent
Describe the job
Read new Outlook mail, keep what needs my attention, and send one WhatsApp briefing every weekday morning.
“Every weekday morning, tell me what actually needs my attention.”
One worker solves a job. A workforce runs the day.
Begin with the work that repeats. Add another worker when the next job is ready to become a reliable operating loop.
Operations
9:00
Unread mail digest
Reads overnight Outlook mail, keeps what needs you, and sends one short WhatsApp summary.
Outlook to WhatsApp
The worker you can take with you.
Your .agent is a readable worker file. Download it, inspect it, share it, customize it, and open it in another supported runtime.
Read the open .agent specRegistered with IANA as application/vnd.agentmug.agent+json
morning-digest.agent
Portable and inspectable
agent.v1Secrets stay outside
API keys, OAuth tokens, and private file paths never live in the .agent file.
Selected .agent layer: Instructions. The worker keeps the instructions, decisions, and guardrails that make the job yours.
Proof before, during, and after every run.
Required sources are checked before work begins. Supported cloud runs remain reviewable after the result is delivered.
Swipe to follow the run
Required source
Permission
Worker run
Delivery
Receipt
Before the model starts
Required knowledge is checked first.
When a worker declares a required source, AgentMug checks that it is present, authorized, current, and still matches the verified revision set.
Bound
Active
Authorized
Current
This applies to workers that declare required sources. AgentMug reads the source material you provide. It does not continuously sync or write back to the provider.
After the work
The result arrives with a receipt.
Status
Duration
Estimated cost
Step-by-step activity
Exact output
Source evidence
Ready is evidence, not decoration.
One definition. Choose where it runs.
Start on AgentMug Cloud, move closer to your machine, call a hosted worker from an AI client, or embed the runtime in your own product.
The simplest place to build and schedule a worker.
Start from a sentence, connect supported accounts, schedule the job, and share the result. Cloud has the full hosted tool catalog and inbound webhooks.
same .agent
Execution
Hosted
Credentials
Connections you authorize
No command line required
Start with a worker that already works.
Add a published worker, connect your own accounts and knowledge, then customize the job for your day.
Published workers are being prepared.
Open the marketplace to see what is available now.
Explore workersFor builders
Build on the same runtime.
Use the .agent specification, TypeScript runtime, CLI, MCP bridge, plugin contract, and OpenTelemetry adapter in systems you control.
Local tools. Your own key. Visible approvals.
Install for Windows, macOS, or Linux. Open a .agent file where your files and shell already live.
WindowsmacOSLinux
Download DesktopDeveloper tools
Install from npm
AgentMug CLI
npm install -g @agentmug/cliTypeScript runtime
npm install @agentmug/runtimeMCP bridge
npx -y @agentmug/mcp-bridgeOpenTelemetry adapter
npm install @agentmug/otelPricing
Start free. Pay when the work scales.
Beta starts automatically with no card. Local and bring-your-own-key runs stay open; paid plans add hosted capacity, managed AI usage, longer history, and support.
No card
to start
BYOK
runs open
Portable
by design
Beta
$0
Build and prove your first useful workers.
- $2 managed AI usage each month
- 3 hosted workers and 1 active schedule
- Unlimited local and bring-your-own-key runs
Scale
$99/month
Higher hosted capacity for a serious operation.
- $50 managed AI usage each month
- 100 hosted workers and 100 active schedules
- 10 concurrent managed runs
Enterprise
Custom security, data residency, deployment, and support terms.
Start with one job. Keep the worker. Grow from there.
Describe what repeats, connect only what it needs, and keep the option to deploy the same .agent somewhere else later.