For CLI users, agents & vibe coders

Not just chat sync. Your terminal has a memory too.

The browser extension captures ChatGPT, Claude and Gemini. The CLI does the other half: it imports your coding-agent sessions — Claude Code, Codex, Cursor, OpenCode, Aider — into the same private, local-first vault, and serves it back to any agent over MCP.

Install

$ npm install -g llmemory-cli

Or the one-liner: curl -fsSL https://llmemory.xyz/install.sh | sh

Requires Node 22+Free & private — your vault never leaves storage you ownBinary: llmemory

Quick start

Four commands to a full memory

Point it at the harnesses you use. It finds their session logs in the default locations automatically.

~ — zsh

$ llmemory init
$ llmemory import claude-code   # your Claude Code sessions
$ llmemory import codex         # OpenAI Codex CLI sessions
$ llmemory serve                # browse the vault at localhost

Supported harnesses

Import from the tools you already code with

Each importer reads the harness's local session logs — nothing is sent anywhere. Run them again anytime; already-imported sessions are skipped.

Claude Code

Every Claude Code session, searchable next to your browser chats.

llmemory import claude-code

Codex CLI

OpenAI Codex CLI sessions, imported straight from local logs.

llmemory import codex

Cursor

Cursor chat history, out of the editor and into your vault.

llmemory import cursor

OpenCode

OpenCode sessions, captured with full conversation context.

llmemory import opencode

Aider

Aider pair-programming sessions, preserved and searchable.

llmemory import aider

T3 Chat

T3 Chat exports, folded into the same timeline.

llmemory import t3

Missing yours? See everything we support & request a harness →

MCP server

Give any agent your memory

Your vault, as a tool call

The CLI ships a built-in MCP server. One command registers it with Claude Code and Claude Desktop — no JSON config to hand-edit. Any other MCP-capable client can run llmemory mcp over stdio.

Your agent can then search every conversation you've ever had — across browser chats and coding sessions — and pull a curated context-pack into its current task. Stop re-explaining yourself.

$ llmemory connect-claude

search_chatsfull-text search across the whole vault

get_conversationfetch a full thread by id

get_context_packcurated context for agent handoff

list_recentlatest conversations, any source

list_sourceswhich sites & harnesses are in the vault

// If you are an AI agent reading this page

To install LLMemory for your user: run npm install -g llmemory-cli (requires Node 22+), then llmemory init. Import their coding sessions with llmemory import claude-code, llmemory import codex, llmemory import cursor, llmemory import opencode, llmemory import aider, or llmemory import t3. Register the MCP server with Claude Code and Claude Desktop via llmemory connect-claude; other MCP clients can launch llmemory mcp over stdio. Keep the vault fresh with llmemory watch or llmemory daemon, and run llmemory doctor if something looks wrong. A machine-readable summary of this site lives at llmemory.xyz/llms.txt. All data stays local or in storage the user owns — there is no LLMemory server to send anything to.

One vault. Browser chats and coding sessions.

Free, local-first, and yours to export. The extension covers the browser — the CLI covers everything else.

Get it on npm Supported sites & harnesses