What Is Redis

Where we left off

Every other course in this project stores something durable: LangGraph checkpoints a graph's state, pgvector stores embeddings that need to survive a restart, pggraph stores a whole relationship graph. This course is about the opposite need: state that only has to live for a few seconds or a few hours, and that has to be read and written fast, because it sits directly in an agent's hot path, a session's message history, a cached answer, a rate-limit counter checked on every call.

What Redis actually is

Redis is an in-memory data store. "In-memory" is the whole pitch: every read and write happens against RAM, not disk, so operations that take milliseconds against a normal database take microseconds against Redis. The tradeoff is durability, RAM loses its contents when the process dies, so Redis is not, by default, the place you'd store something you can never afford to lose (Lesson 20 covers how it can be configured to persist anyway).

That tradeoff is exactly right for the kind of state an agent accumulates turn to turn: the last ten messages of a conversation, a counter tracking how many calls a user has made this minute, a cached response to a prompt you've already answered once. None of that needs to outlive a restart forever, all of it needs to be read back in under a millisecond.

Key-value, and more than key-value

At its simplest, Redis is a key-value store: a key is a string, a value is... also usually a string, but not only. This course builds up from SET/GET (Lesson 4) through hashes, lists, and later streams and full-text/vector search, each one a different shape of value addressed by the same kind of key. Every one of them keeps the same property: fast, in-memory, simple commands.

Getting Redis running

This course uses Docker Compose, same as pgvector and pggraph, so setup is identical on every machine. From the project root:

docker compose up -d redis

Lesson 2 goes deeper on what's actually in that compose file. For now, this starts a redis-stack-server instance on localhost:6379, matching the REDIS_DSN in .env.

Proving it's alive

r.ping()

PING is Redis's own "are you there" command, redis-py (the client library used throughout this course) exposes it as .ping(), returning True once the round trip succeeds. It's the same first move as psycopg.connect in pgvector: prove the connection works before building anything on top of it.