If you want the practical rollout first, start with Quickstart. This page is for the teams that already know they want to wire a live gateway.Documentation Index
Fetch the complete documentation index at: https://docs.atlasmeets.com/llms.txt
Use this file to discover all available pages before exploring further.
The dashboard surfaces
The setup flow is split across three pages.Gateway
Use theGateway page to register the workspace gateway.
Atlas stores:
- manifest URL
- invoke URL
- auth settings
- validation status
- a cached manifest snapshot for UI and debugging
Skills
Use theSkills page to create or edit reusable skill markdown.
Skills live on the Atlas side. They are not part of the customer gateway.
Agents
Use theAgents page to configure the meeting agents your workspace will use.
Each agent controls:
- base realtime prompt
- base brain prompt
- attached skills
- allowed gateway actions
Recommended setup order
- connect the workspace gateway
- validate one read-only action
- create one or two skills
- attach them to the default agent
- trim gateway access down to the actions that agent should actually use
- test in a real meeting
What happens at runtime
When a meeting starts, the worker loads:- the workspace gateway config
- the selected agent config
- the agent’s attached skill metadata
- the filtered gateway action set for that agent
Meeting assistant scope
Before a meeting starts, Atlas can join in one of two listening modes:Room(default)- Atlas listens to the shared meeting audio
- use this for normal multi-participant meetings
Private to host- Atlas listens only to the meeting host
- use this when one person wants Atlas as a private assistant inside a larger room
- use
Roomunless you explicitly want host-only listening
Hosted canary setup
If you are testing Atlas with a hosted worker, the gateway must be reachable from the public internet or private ingress you control. For a quick canary, it is fine to:- run the mock or reference gateway locally
- expose it through a public HTTPS tunnel such as ngrok
- save the tunneled manifest and invoke URLs in the dashboard
- run the sync / async / approval checks from the dashboard first
- then join a real meeting and test the same flows in-room
localhost, 127.0.0.1, or another private-network address, a hosted worker will not be able to reach it.