Loci: shared local memory for agents
Loci
Notes, tasks, contacts and project context live in Markdown on your disk, and Claude Code and Codex read the same memory
- 504 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The installer edits the global instruction files ~/.claude/CLAUDE.md and ~/.codex/AGENTS.md and may add a startup hook
- The storage holds personal data such as contacts, schedule and notes, and the agent sends what it reads to its model
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/lociDetects 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.
- The installer edits the global instruction files ~/.claude/CLAUDE.md and ~/.codex/AGENTS.md and may add a startup hook
- The storage holds personal data such as contacts, schedule and notes, and the agent sends what it reads to its model
This is third-party code. Review the repository files before installing.
What it does
Loci creates a storage folder on disk and connects agents to it through their instruction files. The folder holds Markdown and JSON with personal notes, tasks, a schedule, contacts, links and project state. The agent reads what it needs when a request touches memory or errands and writes new items by the rules in LOCI.md and LOCI-RULES.md. The agent does the installation itself. It reads docs/AI-INSTALL.md from the repository, clones Loci and runs node scripts/loci-install.js. The installer adds a short LOCI.md entry to ~/.claude/CLAUDE.md or ~/.codex/AGENTS.md and, where the client supports it, a hook. Node.js 18 and Git are required; no npm packages are installed. The Dashboard runs as a local Node.js service and shows the same files for viewing and editing. A second agent is connected to the existing storage and sees the same memory.
Who it is for. People who work with several agents and do not want to re-explain their preferences, tasks and project context to each one.
Good fit when
- You have to re-introduce Claude Code to your preferences and current work at every start
- Tasks, meetings and notes are scattered across the chats of different agents, and you want one folder of files you can open and edit by hand
- Codex joins Claude Code and needs the same memory without setup from scratch: the instruction in docs/AI-CONNECT.md is enough
- Project state and technical decisions live in the repository itself, with only a pointer to them in the shared storage
- A local web panel beats asking the agent when you need a quick look at errands and contacts
Not a fit when
- You need a guarantee of execution: rule files steer the model but do not bind it, and the result depends on the client
- You need sync between computers: the storage sits on one disk and the project has no cloud memory service
- The agent works in a plain web chat without access to files and commands: memory is stored on disk and cannot be connected without file access
Example request
Remember this project and tell me where we left off. Remind me tomorrow at 9 a.m. to send the materials.Limitations
The installer connects only Claude Code, Codex and WorkBuddy automatically. For other clients, such as Qoder or Doubao, the agent configures them by hand following docs/AI-CONNECT.md.
The storage is local, but the agent passes data from it to its model under the client's settings, so work does not become fully offline.
The full LOCI-RULES.md manual is read at the first memory operation in a session and takes part of the context.
The project description is mostly in Chinese, with an English version in README.en.md.
How to disable. Remove the Loci block between its markers from ~/.claude/CLAUDE.md or ~/.codex/AGENTS.md and disable the hook if the installer added one. The data folder stays on disk and is moved or deleted by hand. The installer backs up files before changing them.
FAQ
How do I install Loci in Claude Code?
Send the agent the request from the project description: read docs/AI-INSTALL.md and follow it. The agent checks Node.js and Git, clones the repository into a loci folder in your home directory and runs the installer. No npm packages are needed.
How do I connect a second agent to the same storage?
Give it a request pointing to docs/AI-CONNECT.md. The agent finds the storage path in .loci/brain-path in your home folder, adds an entry to its instructions and leaves the data untouched.
Does data go over the network?
The files sit on your disk and the project has no cloud memory service. But whatever the agent reads from the storage goes to its model under the client's usual rules.
What does the storage hold?
Personal notes and decisions, tasks and schedule, contacts and places, saved links with materials and an activity journal. The full project state stays in its repository, and the storage holds only an index.
What does the hook do?
The hook gives the agent short settings at session start. Without a hook the agent calls the same scripts/loci-context.js script per the instruction in LOCI.md, so there is no second rule set.
Discussion
No questions or reviews yet.
Related
WorkflowsSet up Claude Code and Codex in one conversation: global instructions, rules, plugins, skills, a status line and correction memory
A local web panel where your agent builds widgets and screens on top of project files, APIs and MCP servers
The /myoffice command opens a browser office showing what every agent on your machine is doing: thinking, waiting for you or done.
A workflow for Claude Code, Codex and Gemini CLI where the agent files source documents into a persistent, linked wiki
Agent Client for Obsidian
Agent Client
An Obsidian plugin that runs Claude Code, Codex and other agents via the Agent Client Protocol right inside your notes vault
A local Markdown note base with an MCP server and skills: Claude Code, Cursor, Codex and Gemini CLI all read the same context
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Loci: shared local memory for agents (this card) | no | 504 | |
| Claude Code Memory Setup: Obsidian-based memory | no | 1,059 | memory on an Obsidian vault for Claude Code only, with /save and /resume commands |
| Second Brain OS | no | 1,476 | an Obsidian vault template with 18 skills and 72 commands for Claude Code |
README in short
The project is built as one entry, one manual and one data set. A short LOCI.md tells the agent where the storage is, and the full LOCI-RULES.md is read at the first memory operation. Data is loaded on demand, and only short settings are read at start. The repository includes an example storage and is MIT-licensed.