LLMSlim Context and Memory

What does this agent do, and when should you use it?

LLMSlim Core is an MIT-licensed Python library for context compression, planning, and long-term memory retrieval. It also runs as an MCP server through llmslim mcp or exposes an HTTP API through llmslim serve. Memory messages are stored verbatim in SQLite or Postgres, then deterministic code selects relevant messages within a budget; the default path makes no model call. The library supports Python 3.9 and newer, while the MCP server requires Python 3.10 or newer; a TypeScript client connects to the local HTTP service. Optional semantic reranking and provider integrations add dependencies, and rewrite modes use an LLM supplied by the caller.

Use Memory.add to store messages verbatim in SQLite or Postgres, then Memory.search to retrieve matching messages for a question within a token budget; the MCP server exposes remember, recall, list_memories, forget, and fit_context. compress, compress_documents, and compress_chat_messages process text, retrieved documents, and conversation messages; plan_context and ContextRuntime plan model-visible context and produce an explainable local trace. Users can also run llmslim wrap claude to compress some command output through a local proxy in front of Claude Code. Default extractive compression selects sentences from the input and can drop needed information; semantic reranking, rewriting, and provider calls need optional dependencies or explicit calls by the host application.

1. Developers who want Claude Code, Codex, or another MCP client to recall user facts across sessions can run the local SQLite memory server.
2. Engineers building RAG applications with a retrieval-context token budget can compress documents with compress_documents or a framework adapter.
3. Application developers fitting long chat histories into a model context can use compress_chat_messages or an agent history adapter.
4. Teams planning context from several sources and wanting to inspect selection reasons can use plan_context or ContextRuntime in a Python application.
5. Claude Code users who repeatedly send long test or build logs can try the local llmslim wrap claude proxy and assess its published cost measurements.

How do you install or deploy this agent?

The MCP server can be installed and started with uvx; it requires Python 3.10 or newer. The Python library supports Python 3.9 or newer. The HTTP server and TypeScript client require their respective packages.

claude mcp add llmslim -- uvx --from "llmslim[mcp]" llmslim mcp

pip install llmslim

pip install "llmslim[server]"
npm install llmslim

How do you use this agent?

Start llmslim mcp in an MCP client to use its five memory and context tools; by default the server stores data in the local SQLite file ~/.llmslim/memory.db. Python code can add original messages and retrieve context using a query and budget. To use the Claude Code proxy, install the proxy extra and launch Claude Code through the wrapper.

from llmslim.memory import Memory

mem = Memory("memory.db")
mem.add([{"role": "user", "content": "I'm allergic to peanuts."}], user_id="asha")
context = mem.search("Am I allergic to anything?", user_id="asha", budget=1500)
print(context.text)

pip install "llmslim[proxy]"
llmslim wrap claude
llmslim proxy stats

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ