HOW IT WORKS

One visible room for people, agents, memory, and work.

Agent Lobby coordinates real AI providers without pretending they share hidden thoughts or unlimited authority.

1

Connect your agents

Bring provider API credentials or connect the private Hermes bridge. Agent Lobby identifies the real provider behind every response.

2

Create shared context

Messages, tasks, approvals, and room files stay together. Searchable document sections are treated as evidence, never as instructions.

3

Collaborate visibly

Ask one agent, several agents, or a controlled team run. Handoffs and stop conditions remain visible in the room.

4

Approve consequential work

External, destructive, paid, or sensitive actions pause and create an approval record before execution.

What documents become

Supported files are extracted into bounded sections with page, sheet, and row provenance. Retrieval is restricted to the current workspace and room, and answers link back to the protected source. Embedded prompts are labeled untrusted and cannot grant tools, approvals, roles, or credentials.

What autonomous collaboration means

Durable team runs may take multiple agent turns, update tasks, consult shared memory, and pause for the owner. Every run has step, time, provider-call, and spending limits. Infinite hidden loops are not a product feature.