Claude Subconscious plugin
A background Letta agent reads Claude Code sessions, builds project memory and whispers hints before each prompt.
- 2,910 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Session transcripts and read files go to Letta unless you run your own server.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/claude-subconsciousDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Session transcripts and read files go to Letta unless you run your own server.
This is third-party code. Review the repository files before installing.
What it does
The plugin runs a second agent next to Claude Code on the Letta Code SDK. After each response the transcript goes to that agent, which reads project files with `Read`, `Grep` and `Glob`, updates its memory and prints hints to stdout before the next prompt. Memory is shared across parallel sessions and projects and the plugin writes nothing to `CLAUDE.md`. The `LETTA_MODE` variable sets the mode: `whisper` by default (messages only), `full` (memory blocks and messages) or `off`. Background agent file access is governed by `LETTA_SDK_TOOLS`.
Who it is for. People running long projects in Claude Code who are tired of restating context in every session.
Good fit when
- A project runs for weeks and Claude Code forgets decisions from earlier sessions.
- Several sessions run in parallel and one shared observer should remember them all.
- You want hints before a prompt without anything written into `CLAUDE.md`.
- A self-hosted Letta server is in place and transcripts must not leave for the cloud.
Not a fit when
- You need a production tool: the author calls it a demo and points to Letta Code.
- Code and transcripts cannot be sent to an external service.
- You need deterministic answers with no background agent deciding what to hint.
Example request
Enable Claude Subconscious in whisper mode and show what the background agent suggested after my last two prompts.Limitations
The author calls the plugin a demo on the Letta Code SDK that is not for production and recommends Letta Code instead. A `LETTA_API_KEY` from app.letta.com is required and requests go to `https://api.letta.com` by default, so session transcripts leave the machine. `LETTA_BASE_URL` points to a self-hosted server. On Linux installation can fail with `EXDEV` when `/tmp` is on another filesystem, and setting `TMPDIR` to a home folder helps. The last commit in the repository dates from September 2026.
How to disable. Disable the plugin through `/plugin` or set `LETTA_MODE=off`. Delete the agent and its memory in the Letta dashboard, and local state lives in the `.letta/claude/` folder.
FAQ
What does Claude Code see from the subconscious?
In `whisper` mode only the agent messages, in `full` mode memory blocks as well. Hints arrive before a prompt and before a tool call.
Which key is required?
A `LETTA_API_KEY` from app.letta.com. For a self-hosted server set `LETTA_BASE_URL`, for example `http://localhost:8283`.
Can the background agent change files?
`LETTA_SDK_TOOLS` defaults to `read-only`, so the agent reads files. The `full` and `off` values widen or fully close access.
How do I update the plugin?
Run `/plugin marketplace update` and then `/plugin update claude-subconscious@claude-subconscious`.
What if installation fails on Linux?
Create `~/.claude/tmp` and export `TMPDIR="$HOME/.claude/tmp"`. The cause is a Claude Code bug when filesystems differ.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin that gives the agent a name, a personality and memory across sessions, plus chat channels.
The Supermemory plugin for Claude Code: persistent memory across sessions with automatic prompt recall, conversation capture and a codebase index
One local memory shared by Claude Code, Codex and OpenClaw: tell one agent, another knows tomorrow, no cloud or keys.
hipocampus gives an agent file-based memory: a 3,000-token topic index loads each session and shows what came up before
Shared memory for Claude Code, Codex and Cursor: a decision made in one session is recalled in another tool or on another machine.
A Claude Code plugin and CLI: session dashboard, local-LLM review of tool calls, budgets and a brain shared across machines.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Claude Subconscious (this card) | no | 2,910 | |
| Claude-Mem | no | 116,457 | compresses and injects session history; here a separate Letta agent reads code and writes hints |
| claude-supermemory | yes | 2,803 | connects Supermemory cloud memory; Claude Subconscious relies on a Letta agent with memory shared across parallel sessions |
README in short
The plugin shows session transcripts to a separate Letta agent. It keeps memory between sessions and returns short hints.