Skip to main content

​
Synopsis

uzi broadcast <message>
Alias: b

​
Description

The broadcast command sends a message to all active agent sessions in the current repository. The message is delivered to each agent’s tmux pane, simulating user input. This is useful for:
  • Sending the same instruction to multiple agents
  • Requesting status updates from all agents
  • Coordinating parallel work across agents
  • Testing how different agents respond to the same prompt

​
Arguments

​
message
string
required
The message to send to all agents. All arguments are joined with spaces to form the complete message.

​
How It Works

  1. Retrieves all active sessions for the current repository from state
  2. For each session:
    • Sends the message to the agent’s tmux window (:agent)
    • Automatically presses Enter after the message
    • Logs the action

​
Examples

​
Send Same Prompt to All Agents

# Ask all agents to run tests
uzi broadcast "run the test suite and fix any failures"

​
Request Status Updates

# Ask each agent for their progress
uzi broadcast "what have you completed so far?"

​
Multi-word Messages

# No quotes needed - all args are joined
uzi broadcast please review your changes and commit them

# Equivalent to:
uzi broadcast "please review your changes and commit them"

​
Coordinated Actions

# Start multiple agents on related tasks
uzi prompt --agents=claude:3 "waiting for instructions"

# Then broadcast specific instructions to all
uzi broadcast "implement the user authentication flow"

​
Typical Workflow

1

Launch Multiple Agents

uzi prompt --agents=random:3 "standby for task"
2

Verify Agents are Ready

uzi ls
Check that all agents show “ready” status.
3

Broadcast Instructions

uzi broadcast "implement feature X with different approaches"
4

Monitor Progress

uzi ls -w
Watch agents work on the task in real-time.

​
Output

The command displays:
  1. Total number of sessions receiving the message
  2. Each session name as messages are sent
Example output: 
Broadcasting message to 3 agent sessions:

=== agent-myproject-a3f2b1c-thoughtful-tesla ===

=== agent-myproject-a3f2b1c-clever-curie ===

=== agent-myproject-a3f2b1c-brave-bohr ===

​
Use Cases

Launch multiple agents and broadcast the same problem to see different approaches:
uzi prompt --agents=claude:3 "ready"
uzi broadcast "optimize the search algorithm for large datasets"
Have multiple agents test different scenarios:
uzi broadcast "write integration tests for the API endpoints"
Ask all agents to review their work:
uzi broadcast "review your changes, add comments, and commit"
Send stop or reset instructions:
uzi broadcast "stop what you're doing and revert your changes"

​
Behavior Details

​
Message Delivery

  • Messages are sent using tmux send-keys to the :agent window
  • Enter is pressed automatically after the message
  • Double Enter is sent to ensure message submission

​
Session Scope

  • Only affects sessions in the current repository
  • Sessions from other repositories are unaffected
  • Session list is retrieved from Uzi’s state file

​
Error Handling

  • If a session cannot receive the message, an error is logged but other sessions continue
  • No agents found results in an error: no active agent sessions found
  • Missing message argument results in: message argument is required

​
Comparison with run

While both commands execute across all agents, they serve different purposes:
Featurebroadcastrun
PurposeSend messages to agentsExecute shell commands
TargetAgent windowNew tmux window
OutputAgent processes messageCommand output captured
Use CaseAgent instructionsSystem commands
Example: 
# Use broadcast for agent communication
uzi broadcast "implement feature X"

# Use run for shell commands
uzi run "git status"
  • prompt - Create agents to broadcast to
  • ls - See which agents will receive the broadcast
  • run - Execute shell commands across agents
  • auto - Auto-handle prompts triggered by broadcasts

​
Notes

The broadcast command sends messages to all active agents in the repository. Make sure you intend to message all of them before using this command.
Combine with uzi ls to verify which agents are active before broadcasting:
uzi ls && uzi broadcast "your message"
If an agent is in the middle of processing, the broadcast message will be queued and handled when the agent is ready for input.

Build docs developers (and LLMs) love

Get started for freeTalk to us