Trimming and removing messages

Trimming limits what you send the model at call time with trim_messages, while RemoveMessage deletes messages from the saved state for good.

Last updated: 27 Sep, 2026 · LangChain 1.4

A long conversation gets slow and costly to send in full. Trimming controls what the model sees on a call; removing controls what the thread keeps.

Trimming what the model sees

trim_messages returns a shorter list and leaves the saved conversation alone. With token_counter=len it counts messages, so max_tokens is a message budget; here it keeps the last two.

kept = trim_messages(chat, strategy="last",
                     token_counter=len, max_tokens=2)  # keep the last 2

Trimming a four-message chat

The whole snippet, printing the count before and after.

from langchain_core.messages.utils import trim_messages
from langchain.messages import HumanMessage, AIMessage

chat = [HumanMessage("a"), AIMessage("b"), HumanMessage("c"), AIMessage("d")]

# token_counter=len counts messages, so max_tokens is a message budget
kept = trim_messages(chat, strategy="last", token_counter=len, max_tokens=2)

print(len(chat), "->", len(kept))
print([m.content for m in kept])

What trimming did

• The chat had four messages; the model should see fewer.
• strategy="last" kept the most recent two, c and d.
• The original chat list is untouched; only what you pass the model changed.

Removing a message from the saved state

Trimming does not delete anything. To drop a message from a thread for good, return a RemoveMessage with its id from a node; the messages reducer deletes it.

from langchain.messages import RemoveMessage

# in a node: delete every message but the last from the saved thread
return {"messages": [RemoveMessage(id=m.id) for m in state["messages"][:-1]]}

Trim vs remove

When to trim or remove

• Trim when a conversation grows past what the model should read each turn.
• Remove when the saved thread itself should be shortened or reset.
• Summarize instead when older turns still matter but should cost less.

Related

• Previous: Short-term memory with a checkpointer
• Next: Memory that outlives the conversation
• Reference: LangChain docs