letta-ai/claude-subconscious

What it solves

Claude Code typically forgets all context between sessions. Claude Subconscious provides a persistent memory layer and a background intelligence agent that observes your coding sessions, learns your preferences, and maintains project context across different sessions and projects.

How it works

It operates as a plugin for Claude Code using the Letta Code SDK. A background Letta agent monitors session transcripts asynchronously. After a session ends, the transcript is sent to the agent, which can then use tools (like Read, Grep, and Glob) to explore the codebase and update its internal memory blocks. Before each new prompt or tool use, the agent "whispers" relevant guidance or memory updates directly into Claude's context via stdout, without modifying any files on disk.

Who it’s for

Developers using Claude Code who want long-term memory, cross-project consistency, and a proactive background agent that can perform research and pattern detection while they work.

Highlights

• Persistent Memory: Maintains dedicated blocks for user preferences, project context, pending items, and session patterns.
• Background Operation: Runs asynchronously via the Letta Code SDK to ensure it never blocks the main Claude Code workflow.
• Tool Access: The background agent can read files, search the web, and explore the codebase independently.
• Flexible Injection Modes: Offers "whisper" mode for lightweight messages or "full" mode for complete memory block synchronization.
• Multi-Project Shared Brain: A single agent can serve multiple projects, sharing learned knowledge while maintaining separate conversation threads.