It installs as a plugin into the agent CLI you already use, Claude Code or Codex. Everything that makes the agent yours (its character, what it knows about you, your projects and tasks) lives in one folder of plain markdown on your disk. Any model can read it. You can browse it in Obsidian, keep it in git, and take it with you.
Your agent wakes up as Oppi ๐. An octopus keeps many arms busy at once with one brain behind them, and that is how it works for you: several jobs in flight, one memory of all of them. Call it whatever you like; Oppi is just who shows up first. See why an octopus.
On a Mac, your agents live in cmux, and the AgentLayer sidebar sits beside your terminals:
- Manager: every Claude Code and Codex session, the ones waiting on you on top.
- Agents: a chip per agent, and for the one you pick, its Projects (what's due and each project's next task), Reports (pinned, then newest by day) and Goals (today to this year).
The workspace you're in picks the agent. Tasks, reports, the dashboard and the roadmap open as cmux tabs. Setup offers the sidebar when you run it in cmux.
One line starts a setup guide. It checks your machine and shows each step with its link or exact command; you run every command, and it ticks the step when it's done. It never uses sudo.
macOS or Linux:
curl -fsSL https://agentlayer.one/install.sh | bashWindows (PowerShell 7):
irm https://agentlayer.one/install.ps1 | iexIt walks through git, Bun and Claude Code, cmux on a Mac, your agent's folder (~/Documents/Agents/Oppi by default) and the plugin, then hands you the line that starts setup. You can read the guide first. Flags and details: Quickstart.
Rather do it yourself, or starting with Codex? Install by hand has every step, one at a time.
Anyone who would use an agent that remembers, not only engineers. Pick the path that fits you:
| You | Start with |
|---|---|
| Lead a team | Daily rhythm for briefings and goals, then Seed bundles to hand the agent to your team |
| Write code | The engineer skill, then Pull requests, Worktrees and cmux to watch every session at once |
| Run support | Helpdesk for the ticket queue, then Slack and Attention to see who is waiting on you |
| Want a personal assistant | Your first ten minutes, then Daily rhythm and Capture |
flowchart LR
S["<b>Sessions</b><br/>Every conversation,<br/>saved when it ends"]
K["<b>Knowledge</b><br/>Compiled into notes it<br/>reads before you type"]
P["<b>Projects</b><br/>Tasks and plans<br/>it keeps current"]
R["<b>Results</b><br/>Briefs, reviews<br/>and shipped work"]
S -->|capture| K
K -->|informs| P
P -->|generate| R
R -->|feeds the next session| S
Every session is saved when you exit. Saved sessions are compiled into notes the agent keeps. Those notes load before you type a word in the next session. Everything else is built around that loop: skills and tools feed it and act on what it knows. โ How the brain works
- You drive. It never starts a session on its own. Ask in plain words and it picks the skill; you never have to remember a command.
- No metered API key. Its tools return prompts and your own session does the thinking, on the plan you already have.
- Local by default. No cloud sync unless you put the folder somewhere that syncs. Keys sit in a store the agent's own tools cannot read. See Privacy.
- One plugin, many agents. A personal agent and a work agent share one install and keep separate folders. Always launch from the agent's folder: that's what loads the plugin and keeps agents apart.
- 34 skills with 74 playbooks: briefings, goals, attention and standup, engineering and PR review, sync, tax, helpdesk, trips, SEO, and more.
- 83 MCP tools for tasks, knowledge, reports, history, worktrees, GitHub, databases, a bundled browser and search.
- Four hooks that load the brain at session start and capture the session at the end.
- Opt-in packs: SEO, Browser, Database, GitHub, API, Xcode, Helpdesk, App Store, Slack and cmux.
- A dashboard, one static HTML page of your agent's day, work and brain, regenerated on every sync.
Everything lives at agentlayer.one/docs, also served as llms-full.txt for models.
| Section | Start with |
|---|---|
| Start here | Quickstart ยท Your first ten minutes ยท How it works ยท What setup asks ยท Install by hand |
| Use it | Daily rhythm ยท Attention ยท Tasks ยท Capture ยท Sync |
| Workstation | The rig ยท Terminal ยท cmux ยท Editor and tools |
| Engineering | The engineer skill ยท Pull requests ยท Second-model review ยท Worktrees ยท Releases |
| Connect it | GitHub ยท Database ยท Browser ยท SEO ยท Slack ยท Helpdesk ยท App Store ยท Xcode ยท API |
| Run it | Updating ยท Hosts ยท Configuration ยท Adding another agent ยท History ยท Privacy |
| Reference | How the brain works ยท Agent home ยท Skills ยท MCP tools ยท CLI ยท Hooks ยท Changelog |
| About | FAQ ยท Platforms and limits ยท Contributing |
Pull the new plugin version on your host (/plugin update agentlayer@agentlayer in Claude Code; remove and re-add in Codex), then run the upgrade skill. The plugin update refreshes code; upgrade brings your folder in line, backing up first. On a version from before 1.0, your home moves to a new folder instead. โ Updating
Pull requests welcome: new opt-in packs, read-mostly MCP tools, platform hardening, docs. Open an issue first for architectural changes; the plugin's contract with the markdown home is intentional. See CONTRIBUTING.md.
Licensed under the Apache License, Version 2.0. agentlayer is ยฉ AgentLayer ยท agentlayer.one. See NOTICE for the attribution stanza. Third-party skill libraries installed via the configure-skills skill are not bundled and carry their own licenses.
Built by AgentLayer ยท An agent that remembers. In a folder you own.
