Ch 2 — Your First Chain Under the Hood

What actually happens when you call chain.invoke() — every object, every transformation
Index ← High Level
Under the Hood
-
Click play or press Space to begin...
Step - / 10
A The Input Dictionary What .invoke() receives
1
data_object
Input Dict {"topic":
"quantum computing"}
passed to
link
RunnableSequence .invoke() starts
the pipeline
step 1
edit_note
prompt.invoke() First Runnable
in the sequence
2
arrow_downward Prompt formats dict into Message objects
B Inside ChatPromptTemplate.invoke() dict → ChatPromptValue
find_replace
Variable Sub {topic} →
"quantum computing"
creates
person
SystemMessage "You are a
helpful assistant."
+
chat
HumanMessage "Explain quantum
computing in 1 sentence"
wraps
3
inventory_2
ChatPromptValue Wrapper holding
the message list
4
arrow_downward ChatPromptValue flows into model.invoke()
C Inside ChatOpenAI.invoke() ChatPromptValue → HTTP request → AIMessage
transform
Serialize Messages → OpenAI
API JSON format
POST
cloud
HTTP Request POST /v1/chat/
completions
response
5
download
Parse Response JSON → AIMessage
with metadata
6
arrow_downward The AIMessage object in detail
D The AIMessage Object What the model actually returns
text_fields
.content The response
text string
+
info
.response_metadata Model name, tokens,
finish reason
+
build
.tool_calls Empty list (no
tools in this chain)
+
tag
.id Unique run ID
for tracing
7
arrow_downward AIMessage flows into parser.invoke()
E Inside StrOutputParser.invoke() AIMessage → plain string
psychology
AIMessage Full object with
content + metadata
.content
text_fields
Extract String return
message.content
=
check_circle
Plain String "Quantum computing
uses qubits..."
8
arrow_downward Now let's see how .stream() works differently
F Streaming: .stream() vs .invoke() Token-by-token output
play_arrow
prompt.invoke() Runs fully
(not streamable)
all at once
9
stream
model.stream() Yields AIMessage
Chunks one by one
chunks
output
parser.stream() Passes each
chunk through
10
arrow_downward Callbacks fire at every step for tracing
G Callback Events What fires during chain execution
play_circle
on_chain_start Fires when
chain begins
then
smart_toy
on_llm_start Fires when model
call begins
then
stream
on_llm_new_token Fires per token
(if streaming)
finally
stop_circle
on_chain_end Fires with
final output
1
Detail