Multiplayer Threads

Sazabi supports multiplayer debugging, where multiple team members can participate in the same thread simultaneously. Messages sync in real time across all clients, making collaborative investigations seamless.

How multiplayer works

When you open a thread, you see the same state as everyone else viewing it. This includes:

• All messages in the conversation
• AI responses as they stream in
• Tool calls and their results
• Artifacts like charts and tables

Any team member can send a message to the thread. When someone sends a message, all viewers see it appear immediately. When the assistant responds, everyone sees the response stream in real time, even if they were not the one who asked the question.

Joining an active thread

There are several ways to join a thread that a teammate is working on.

From the dashboard sidebar

Organization-wide threads appear in the sidebar for all team members. If a teammate is actively investigating an issue, you can click the thread to join.

Look for threads with recent activity. The sidebar shows the most recent message preview and timestamp, so active threads appear near the top.

From a shared link

Team members can share direct links to threads. Click a shared link to open the thread and join the conversation. You will see the full history and any ongoing activity.

From Slack

If your organization uses the Slack integration, threads started in Slack are synced to the dashboard. Click the link in a Slack thread to open it in the dashboard, or find it in the sidebar.

Real-time message sync

Messages sync across all connected clients within milliseconds. The sync includes:

Streaming responses

When the assistant generates a response, all viewers see it stream in real time. This works regardless of who sent the message that triggered the response.

Who is viewing a thread

Sazabi currently focuses on message sync rather than live presence indicators. You do not see avatars or cursors showing who else is viewing the thread in real time.

However, you can identify who has participated in a thread:

• Message attribution: Each user message shows who sent it
• Timestamps: See when each message was sent
• Thread history: The full conversation shows all participants

Concurrent interactions

Multiple team members can interact with the same thread simultaneously. Here is how different scenarios are handled.

Multiple messages in sequence

If two people send messages at nearly the same time:

1. Both messages are queued in the order they were received
2. The assistant processes them sequentially
3. Each message gets its own response
4. All viewers see messages and responses in the correct order

The thread maintains a consistent order across all clients.

Interacting with a streaming response

While the assistant is responding to one question, other viewers can still read the conversation, expand previous tool calls, and interact with existing artifacts. New messages are queued until the current response completes.

Interrupted responses

If a response is interrupted (for example, due to a network issue), all viewers see the same interrupted state. The message shows as failed and can be retried by any team member with access.

Use cases for collaborative debugging

Multiplayer threads are particularly useful for these scenarios.

Incident response

During an outage, multiple engineers can join the same investigation thread:

• One person describes symptoms while another searches for related alerts
• Everyone sees the assistant's findings as they stream in
• The team builds shared context without repeating questions
• Later, the full thread serves as an incident timeline

Onboarding and knowledge transfer

Senior engineers can guide newer team members through investigations:

• The senior engineer asks questions and explains the reasoning
• The newer team member follows along and asks follow-ups
• Both see the same evidence and can discuss findings
• The thread becomes a learning resource

Cross-team investigations

When an issue spans multiple services owned by different teams:

• Each team joins the thread with their domain expertise
• The assistant pulls data from all connected sources
• Teams see findings from services they do not own
• The conversation captures the full cross-team investigation

Pair debugging

Two engineers working on the same problem:

• One person drives while the other observes and suggests directions
• Both see tool outputs and can spot patterns
• They can switch who is asking questions at any time
• No need to share screens or relay information manually

Thread visibility and access

Multiplayer only works for threads that team members can access.

• Organization-wide threads: Any member of the organization can join
• Private threads: Only the creator can see the thread

For organization-wide threads, all members see the same thread in their sidebar and can join at any time. There is no invitation step.

See Thread Visibility for more about controlling who can access your threads.

Limitations

Current limitations of multiplayer threads:

• No presence indicators: You cannot see who else is currently viewing
• No collaborative editing: Each message is sent by one person; there is no concurrent draft editing
• Sequential processing: The assistant handles one message at a time per thread

These limitations reflect the current focus on reliable message sync. The goal is to ensure everyone sees the same conversation state, not to enable simultaneous editing of the same message.