Skip to main content

Conversations

A conversation is the container for an ongoing dialogue between a user and an agent. Think of it like a chat thread -- it keeps track of everything that has been said so the agent can maintain context across multiple messages.

How Conversations Work

Every conversation:

  • Maintains full context across messages so the agent remembers what was said earlier
  • Has a unique identifier you can use to resume it later
  • Stores history and metadata for analytics and audit
  • Is scoped to a workspace, so the agent only sees relevant resources

Conversation Lifecycle

Conversations move through a simple set of states:

StateWhat It Means
ActiveCurrently in use
CompletedFinished -- outcome recorded (resolved, escalated, etc.)
ArchivedHistorical record, no longer active

Conversations stay Active while in use and move to Completed when explicitly ended or resolved. You can archive conversations manually from the dashboard for long-term record keeping.

Workspace Scope

Conversations carry workspace context. When you start a conversation while a workspace is selected in the dashboard, the conversation is automatically scoped to that workspace:

  • The agent's tools only see resources within that workspace
  • Any resources created during the conversation go in the correct workspace
  • Loyd confirms placement before creating resources

This context is set when the conversation is created -- changing the workspace selector mid-conversation does not retroactively change the scope.

Context Window Management

Long conversations are automatically managed so the agent always has the most relevant context:

SituationWhat Happens
Under token limitFull history sent to the model
Approaching limitOlder messages summarized automatically
At limitSummary of earlier messages plus full recent messages

File Attachments

You can attach files directly in chat conversations. Files are uploaded, security-scanned, auto-parsed, and their content is injected into the agent's prompt.

Click the paperclip icon next to the message input to attach files. This works in Admin conversations, Openspace chats (team, agent, and Loyd), and the AG-UI stream. Up to 10 files per message.

When you send a message with attachments, MeetLoyd automatically:

  1. Resolves the parsed content of each file
  2. Injects the full text into the agent's prompt (token-budgeted per model)
  3. Stores a compact summary in conversation history (not the full content)

The agent can immediately read and reference your documents without downloading them. See Files and Attachments for full details.

Live Message Injection

You can send messages while an agent is still responding -- for corrections, additional context, or to change direction mid-task. The chat input stays enabled during streaming. Messages sent mid-stream appear instantly in the UI and are picked up by the agent at its next processing boundary.

Multi-Agent Conversations

Escalation

When an agent escalates, the conversation continues seamlessly. The new agent receives the full conversation history and a summary of what has been tried.

Handoff

Agents can transfer ownership of a conversation to another agent. The calling agent stops, the target agent picks up with full context. This is the core collaboration primitive in MeetLoyd.

Managing Conversations

Starting a Conversation

Click on any agent in the dashboard or Openspace to start a new conversation. You can also start conversations with an entire team -- the message goes to the team's entry agent, who routes it as needed.

Viewing History

Go to Conversations in the dashboard to see all conversations. You can filter by agent, team, status, and date range.

Ending a Conversation

Conversations can be explicitly ended from the dashboard. Ending a conversation records the outcome (resolved, escalated, etc.) which feeds into analytics.

Conversation Metadata

Conversations support metadata tags (ticket ID, priority, department, source) that help with filtering, analytics, and audit. Metadata is set automatically from context or can be added manually.

Agent-Generated Documents

Agents can create documents during conversations using the generate_document tool. Generated files appear as artifact cards in the chat with format-specific icons and a download button. See Document Generation for details.


Files & Attachments
Upload documents and let agents read them
Memory
Persistent context across conversations
Streaming & Real-Time
Live message injection and streaming details