Streaming Mode

Logician streams everything — reasoning, tool calls, and results — in real time. You see the agent think before it acts.

How it works

Stream stages

Stage	Events	Description
Turn start	turn_start, message_start	New turn begins
Thinking	thinking_token	Agent's internal thought process
Response	token, message_update	Streaming text output
Tool execution	tool_execution_start, tool_execution_update, tool_execution_end	Tool invocation with progress
Turn end	turn_end, agent_settled	Turn completes, agent settles

Configuration

Streaming is enabled by default. Control it via config:

{
  "streaming": {
    "enabled": true,
    "showThinking": true,
    "showToolCalls": true,
    "showToolResults": true
  }
}

Headless streaming

In headless mode, streaming outputs as JSONL:

npm start -- exec --jsonl "analyze src/auth.ts"

Each line is a typed JSON event:

{"type":"turn_start","turn_id":"turn_1"}
{"type":"thinking_token","token":"Analyzing auth flow..."}
{"type":"tool_execution_start","tool":"read_file","tool_name":"read_file","tool_call_id":"tc_1","tool_args":{"path":"src/auth.ts"}}
{"type":"tool_execution_end","tool":"read_file","tool_name":"read_file","tool_call_id":"tc_1","result":"...","is_error":false}
{"type":"message_update","turnId":"turn_1","message":{"role":"assistant","content":"Found 3 auth middleware functions..."}}
{"type":"turn_end","turn_id":"turn_1","message":""}