Agent Builder Chat & Collaboration Enhancements

"**Value proposition**

Teams increasingly collaborate around agent conversations rather than working with agents solo. This work improves how conversations are organized, attributed, and started, including structured metadata, a smoother first-run experience for new users, and multi-user active chat.

**Expected outcome**

- Structured metadata attached to conversations, plus reusable metadata templates that allow conversations to be shareable assets for collaboration.
- Suggested conversation starters and quick-select suggestion chips to help new users engage faster.
- Multi-user chats allowing users to collaborate with each other and the agent in a single conversation to drive investigation, incident response, research and more.
- Agents can be assigned to spaces to allow for collaboration across teams organized by space

**Key user stories**

- As a team, I want to be able to collaborate on an investigation and use the agent as a team member to help with tasks and respond to asks. I want all messages to be properly attributed and access controls maintained.
- As an admin, I want to tag conversations with structured metadata (e.g., case ID, customer) so they're easier to find and report on.
- As a new user, I want suggested starting points so I know what to ask an agent.
- As an admin, I want to enable teams across my org to access agents specifically for their data. These teams are separated by space so I want to ensure each space has a configured default Agent.

**Phasing**

- Phase 1 (Next): Conversation metadata & templates, sharable conversations, prompt suggestions chips, agent per space
- Phase 2 (Later): Active Multi-user chat

The issue does not identify repository files, tests, or entry points to begin with. First locate the existing Agent Builder conversation, space, and access-control components, then clarify the phase-one scope and acceptance criteria for metadata, templates, suggestion chips, and a default agent per space; done requires those outcomes to work together without weakening attribution or access controls.

Written by the indexing model from the issue text.