Rlm Infinite Memory For Claude Code

What is Rlm?

Rlm is an MCP server that gives Claude Code persistent memory across sessions. It solves the problem of context window limits and the loss of conversation history after /compact, allowing users to save and recall decisions, facts, and conversation segments. It is designed for Claude Code users who need long-term memory.

How to use Rlm?

Install Rlm via pip, uv, Git, or Docker (requires Python 3.10+ and Claude Code CLI). After installation, Rlm is automatically configured with zero additional setup. Use its 14 tools, such as rlm_remember() and rlm_recall(), to save and retrieve insights; rlm_chunk() and rlm_peek() for conversation history; and rlm_search() for hybrid search. Hooks auto-save before context loss.

Key features of Rlm

• 14 tools for memory and conversation management
• Auto-save before /compact or auto-compact (PreCompact hook)
• Two memory systems: Insights (key decisions) and Chunks (full conversation)
• Multi-project organization with auto-detection from git/working directory
• Smart retention with 3-zone lifecycle (Active, Archive, Purge)
• Hybrid semantic search (BM25 + cosine similarity, optional)
• Sub-agent skills for chunk analysis (/rlm-analyze, /rlm-parallel)

Use cases of Rlm

• Remembering client preferences and project decisions across sessions
• Recalling past architectural choices and findings
• Organizing conversation history by project or domain
• Searching for specific content with regex, fuzzy, or semantic queries
• Automatically saving context before a session is compacted

FAQ from Rlm

How does Rlm differ from Letta/MemGPT?

Rlm works natively with Claude Code via MCP, while Letta/MemGPT requires its own runtime. Rlm offers auto-save before compact, hybrid search (regex + BM25 + semantic), fuzzy search, multi-project support, smart retention, and sub-agent analysis—all with zero-config installation.

What are the installation requirements?

Rlm requires Python 3.10+ and the Claude Code CLI. It can be installed via pip install mcp-rlm-server[all], uv, Git, or Docker.

Where does Rlm store data?

Data is stored in ~/.claude/rlm/ by default. This includes session memory (insights), chunk index, conversation chunks, compressed archives, and optional semantic embeddings.

Does Rlm automatically save conversation history?

Yes. Rlm hooks into Claude Code’s /compact event (PreCompact hook) to automatically save a snapshot of the current conversation before context is wiped. No manual action is needed.

What search capabilities does Rlm offer?

Rlm provides regex search (rlm_grep), BM25 keyword search (rlm_search), and optional hybrid semantic search (BM25 + cosine similarity) using Model2Vec (fast, default) or FastEmbed (more accurate). Fuzzy matching is available for typo-tolerant searches.