The workspace
Agents
An agent is a reusable assistant with a name, a set of tools and a ceiling on what it may do without asking you first. Install one from the Marketplace, or build your own.
Every action passes the Trust Layer
Allowed actions run, risky ones pause for your approval, and forbidden ones are blocked before anything happens rather than reported afterwards.
Mention one from any chat
Type the at sign and an installed agent joins that conversation, bringing its own tools and instructions with it.
You can read the run
Every dispatch shows the plan it chose, which model handled each step, and what each step produced.
How it works
- 01
Install or build
Take one from the Marketplace, or describe the job and let Alias draft the agent.
- 02
Set the ceiling
Choose which tools it may reach and how far it may go before it has to ask you.
- 03
Put it to work
Mention it in a chat, or dispatch it on a job and read the run when it finishes.
Try asking
Build me an agent that turns support tickets into a weekly summary.
Ask my code review agent to look at the open pull requests.
Set up an agent that watches a competitor's pages and reports what changed.
Questions
- Can an agent do something I did not approve?
- No. Every tool call is checked against the ceiling you set, and anything above it stops and waits for you instead of proceeding and telling you about it later.
- Can I author agents on a personal account?
- Personal accounts install curated agents from the Marketplace rather than authoring their own. An installed agent is still entirely yours and answers in your chats.
- Which model does an agent use?
- Whichever suits each step. An agent that needs tool calling is never routed to a model that cannot make them, so a plan does not fail halfway through on a technicality.