stream message

This block is part of the anthropic group and was last modified in core v15.0.0.

stream message anthropicconnection v15.0.0

Send a prompt to Claude and stream the reply token-by-token to a WebSocket channel as it generates, then return the full text.

In pins 12 + flow

flow
flow
connection required
anthropicconnection
memory
anthropicmemory
Optional. A memory (from create memory); prior turns are loaded in and this exchange is appended, so the chat continues across runs.
prompt required
value
The user prompt.
system
value
Optional system prompt (instructions / persona).
attachments
file
Optional images/PDFs to send with the prompt. For large documents with an expensive model, run them through preflight document first to cut tokens.
model
value
Model ID. Defaults to the connection's default model. • Suggestions: claude-opus-4-8 claude-sonnet-4-6 claude-haiku-4-5
max tokens
number
Max tokens to generate. Defaults to 16000.
thinking
condition
Enable adaptive extended thinking.
cache
condition
Cache the system prompt (prompt caching) for cheaper repeated calls.
effort
value
Reasoning effort: low | medium | high | xhigh | max. • Suggestions: low medium high xhigh max
channel
value
WebSocket channel to broadcast tokens to (the frontend subscribes to it). Leave empty to stream straight to the calling WebSocket connection instead.
event
value
Event name for the broadcast frames. Defaults to anthropic_delta.

Out pins 11 + flow

flow
flow
connection
anthropicconnection
memory
anthropicmemory
The memory, passed through with this turn appended; chain it into the next send message.
success
condition
text
value
The full reply (also delivered live over the channel/spark).
usage
object
Token counts plus the call cost: input_tokens, output_tokens, cache_read_input_tokens, cache_creation_input_tokens, cost_usd.
stop reason
value
partial
value
errors
array<value>
Error codes when the call fails (success=false): config_error, internal_error, model_not_allowed, budget_exceeded, concurrency_limit_exceeded, rate_limited, refused, max_tokens, transport_error.
warnings
array<value>
Non-fatal notices: cost_unknown.
error detail
value
on delta
execute_function
Optional: run this function for every streamed chunk. It receives the chunk, its index and the text so far. Keep it light; it runs synchronously per chunk and a heavy callback slows the stream.

stream

live message

streaming chat

typewriter

sse

A typical wiring for stream message: a function trigger starts the flow; value feeds the prompt pin; the function out pin wires to an execute block.

Studio canvas example for the stream message block: typical wiring for stream message.

Introduced in v15.0.0.

Cluster