Lists for Conversation History

Where we left off

A hash (Lesson 6) is great for a session's metadata, a few named fields. A conversation isn't that shape, it's an ordered sequence of messages that grows one at a time. Redis has a value type for that too.

Lists: an ordered, growable sequence

r.rpush("session:abc123:messages", "user: what's the weather in Paris?")
r.rpush("session:abc123:messages", "assistant: 18C and partly cloudy.")

A list is an ordered sequence of strings under one key. RPUSH key value appends to the right (tail) end, the natural direction for a message log, each new message is pushed after the ones before it. (LPUSH pushes to the left/head instead, useful when a list is being used as a queue read from the other end, see Lesson 15.)

Reading a range

history = r.lrange("session:abc123:messages", 0, -1)
last_two = r.lrange("session:abc123:messages", -2, -1)

LRANGE key start stop returns a slice, using the same negative-index convention as Python: 0 is the first element, -1 is the last. 0, -1 reads the whole list; -2, -1 reads just the last two, the kind of call an agent makes to pull only recent context instead of the entire history.

Keeping a history bounded

r.ltrim("session:abc123:messages", -50, -1)

An unbounded chat log grows forever, and most of it stops being useful context after a while. LTRIM key start stop keeps only the range given and discards the rest, in place. Calling LTRIM key -50 -1 after every push keeps a list capped at its most recent 50 entries, without a separate cleanup step.

Length, without reading the whole list

r.llen("session:abc123:messages")

LLEN returns a count in constant time, no need to LRANGE the whole list just to check how many messages exist.