Published by Dahshan Labs

AI workers for recurring business tasks

AgentMug is a platform by Dahshan Labs for building and running portable AI workers. Give a worker a specific job, connect the accounts and knowledge it needs, and review its runs. Business users can start in the web app; developers can also work with the .agent file, runtime, and CLI.

What is an AI worker?

An AI worker is a saved job definition with instructions, tools, and requirements. It can use an AI model to interpret a request and call the tools available to it. A useful starting job has a clear input, a bounded action, and a result you can check.

Related documentation

Choose one task with a clear result

For example, start with a research brief from specified sources, a draft response for review, or a recurring summary. These are starting points for a worker you configure, not promises that every account already has the required tools. Define the destination, the schedule if needed, and what counts as a successful result.

Related documentation

Connect accounts and knowledge

Create a worker from a description or customize one from the marketplace. Review its proposed tools and connect the required accounts. A connection does not imply that every action offered by that service is supported. Check the worker's requirements and resolve missing permissions or sources before running it.

Related documentation

Test before repeating the work

Use a small, representative input. Inspect the output and the run history, check the facts against the source, and confirm any external change in the destination service. Adjust the instructions and repeat the test before relying on a schedule. Model output can contain mistakes; keep human review where the result matters.

Related documentation

Keep the worker definition portable

Export the worker as a .agent file to share its job definition. Credentials are configured separately by each user and runtime. Cloud, desktop, and CLI have different tool support, so check the destination's requirements before moving a worker.

Related documentation

A first task you can test

Example to try with your own sources.

Start with a draft research brief. Supply three public source URLs and the date range you care about. Configure the tools needed to read those sources before testing.

Example request

Read the three source links I provide. Return one useful update per source, with its URL and publication date. Name any source you could not verify. Return the draft here; do not send or publish it.

What to check before scheduling

  • Each factual update links to one of the supplied sources.
  • Missing, inaccessible or undated sources are identified, not filled in with guesses.
  • The output stays a draft; check that no external message or publication occurred.
  • Repeat the manual test with changed sources before relying on a schedule.

Frequently asked questions

Do I need to write code?
You can describe a job and configure a worker in AgentMug's web app. Some jobs require account setup, API keys, or technical configuration. Developers can author and validate the same .agent format in code.
How is this different from a chatbot?
A chatbot primarily provides a conversation. An AgentMug worker stores a reusable job definition and can use configured tools and connections to carry out that job. Its available actions depend on the runtime, permissions, and tools you set up.
Does a shared worker include my credentials?
The portable format declares connection requirements separately from credentials. The person importing the worker supplies their own connections. Review instructions and embedded content before sharing a file; do not put secrets in free-text fields.
What does it cost?
AgentMug Cloud uses managed model inference under its current plans and limits. Desktop and local CLI runs use your own configured model access; provider charges and connected-service costs are separate. Check pricing and test usage before expanding a recurring workflow.

References and next steps

Continue with the other guide: Build and deploy a portable AI agent