langchain-collapse 0.1.0

Collapse consecutive read/search tool calls to reduce agent context size

langchain-collapse

Collapse consecutive read/search tool-call groups to save context.

Quick Install

pip install langchain-collapse

🤔 What is this?

Most context management is reactive — it compresses after the context window is already full. This middleware is preventive. It collapses consecutive read/search tool-call groups into a short placeholder before they accumulate, so expensive LLM-based summarization triggers less often.

No LLM calls. No hallucination risk. Stateless.

langchain-collapse benchmark
==================================================
                               Before    After
                    Messages       37        9
               Approx tokens     2531      197
             Token reduction               92%

SummarizationMiddleware trigger point (at 85% of 200K):
  Without CollapseMiddleware:  ~1214 tool calls
  With CollapseMiddleware:     ~5151 tool calls
  Ratio:                       4.2x more work before summarization

Usage

from langchain.agents import create_agent
from langchain_collapse import CollapseMiddleware

agent = create_agent(
    model="anthropic:claude-sonnet-4-6",
    tools=[...],
    middleware=[CollapseMiddleware()],
)

Place before SummarizationMiddleware so it reduces message count first:

from langchain.agents.middleware import SummarizationMiddleware

middleware = [
    CollapseMiddleware(),
    SummarizationMiddleware(
        model="anthropic:claude-haiku-4-5-20251001",
        trigger=("fraction", 0.85),
    ),
]

How it works

Five consecutive read_file calls (10 messages) become 3:

Before:
  AIMessage(read_file "auth.py")     ─┐
  ToolMessage(content)                │  collapsed
  AIMessage(read_file "users.py")     │
  ToolMessage(content)                │
  AIMessage(read_file "db.py")        │
  ToolMessage(content)                │
  AIMessage(read_file "routes.py")   ─┘
  ToolMessage(content)
  AIMessage(read_file "config.py")   ─── kept (most recent)
  ToolMessage(content)               ─── kept

After:
  HumanMessage("[4 tool results omitted — most recent result preserved below]")
  AIMessage(read_file "config.py")
  ToolMessage(content)

Configuration

CollapseMiddleware(
    collapse_tools=frozenset({"read_file", "grep", "glob", "web_search"}),  # default
    min_group_size=2,  # minimum consecutive pairs to collapse
)

📖 Development

git clone https://github.com/johanity/langchain-collapse.git
cd langchain-collapse
pip install -e ".[test]"
pytest

📕 License

MIT