Ch 1: Tech Stack Overview — Under the Hood

Ch 1 — Tech Stack Under the Hood

Code-level internals: Runnable protocol, StateGraph, Pregel engine, event loops, tracing

Index ← High Level

Under the Hood

Click play or press Space to begin...

Step - / 10

A The Runnable Protocol LangChain's universal interface

api

Runnable Base class for
every component

.invoke()

play_arrow

.invoke() Single input
single output

.stream()

stream

.stream() Yields chunks
as they arrive

.batch()

dynamic_feed

.batch() Parallel execution
of multiple inputs

arrow_downward Runnables compose via the | pipe operator

B RunnableSequence & Composition How the | pipe operator works

edit_note

Prompt Runnable that
formats messages

smart_toy

Model Runnable that
calls the LLM

output

Parser Runnable that
extracts output

link

RunnableSequence Also a Runnable!
Recursive composition

arrow_downward LangGraph: StateGraph definition

C StateGraph Definition Nodes, edges, and typed state

data_object

TypedDict State Schema for the
graph's memory

used by

hub

add_node() Register a
function as node

wired by

call_split

add_edge() Static or
conditional routing

compiles

compiled

.compile() Produces a
CompiledGraph

arrow_downward Under the hood: the Pregel execution engine

D Pregel Execution Engine How CompiledGraph actually runs

inbox

Channels State fields as
message channels

read

hub

Nodes Execute Read channels,
run function

write

edit

State Update Reducer merges
partial updates

check

route

Edge Check Evaluate edges,
pick next node

arrow_downward The execution loop and checkpointing

E The Execution Loop Read state → execute node → write state → repeat

visibility

Read State Snapshot current
state from channels

run

psychology

Execute Node LLM call happens
HERE inside node

save

Checkpoint Persist state
for resumability

loop?

loop

Continue? Conditional edge
decides: loop or end

arrow_downward Chainlit: event wiring and callbacks

F Chainlit Event Wiring Decorators, messages, streaming callbacks

alternate_email

@cl.on_message Decorator registers
message handler

creates

chat

cl.Message Object with content,
author, elements

streams

stream

stream_token() Pushes tokens
to browser via WS

shows

account_tree

cl.Step() Nested steps for
tool calls, reasoning

arrow_downward LangSmith: how tracing hooks into every Runnable

G LangSmith Tracing Internals Callbacks, runs, and trace structure

settings

Env Variable LANGCHAIN_
TRACING_V2=true

activates

webhook

CallbackHandler Wraps every
Runnable execution

emits

account_tree

Run Tree Parent → child
run hierarchy

captures

analytics

Trace Data Inputs, outputs,
latency, tokens, $