Multi-Agent Workspace

Summary

Multi-Agent Workspace is the future model where scoped agents can collaborate inside a workspace while remaining visible and permissioned.

Current implementation status is UI placeholder only inside Alma. No backend agent orchestration exists.

Agent Requirements

Every agent should be:

• observable
• pausable
• permissioned
• auditable
• scoped to a workspace or app domain

Example Agent Roles

The UI placeholder uses abstract roles:

• Research Agent
• Builder Agent
• Ops Agent
• Finance Agent
• Product Agent

These roles are not backend workers yet. They represent future task lanes.

Workspace Scope

Agents should inherit:

• workspace context
• allowed apps
• allowed data scopes
• artifact output policy
• confirmation policy
• audit destination

Approval Model

Agents can plan, summarize, and prepare artifacts. Medium-impact and high-impact actions should stop at approval prompts.

Non-goals In Current Task

• No queue.
• No cron.
• No long-running worker.
• No provider adapter.
• No real multi-agent backend.