Install in one line
The package ships its own runtime — no Bun, no Node, no Docker required. The same command installs and updates.
Self-hosted · MCP native · Apache-2.0
Kernl
Agent teams that do your work,
on your own machine.
A self-hosted server that hands any LLM your real data — tasks, mail, calendar, finance, contacts — and runs teams of agents that act on it. No account, no cloud, nothing leaves the box.
$ curl -fsSL https://lifekernl.com/install.sh | sh
Debian/Ubuntu · Fedora/RHEL · macOS · Windows (.msi) → dashboard on localhost:3086
103 tools on install
71 bundled modules
290+ tools when enabled
3086 dashboard port
The problem
Your AI is brilliant
and amnesiac.
Every conversation starts from zero. Your LLM can't see your tasks, read your inbox, check your budget, or remember what you told it yesterday.
And the cloud assistants that promise to fix that want your entire life uploaded to their servers. Kernl fixes both, without asking you for either concession.
Architecture
One endpoint, all your tools
Getting started
Four steps, no terminal after the first
Pick your LLM
First run walks you through four steps. Bring a cloud key, or keep everything local with LM Studio or Ollama.
Connect any MCP client
Claude Desktop, Cursor, or your own. stdio and Streamable HTTP are served at the same time on /mcp.
Put the agents to work
Spin up an office and let a team run in loops while you are away. Watch them in the live 3D office view.
Capabilities
What's inside
71 modules
Tasks, contacts, reminders, email (IMAP/SMTP), finance, calendar, notes, goals, health, shopping, travel and documents — every one exposed as MCP tools.
Agent offices
Not a chatbot. Cooperating agents with chains, schedules and their own workspaces, packaged as installable .kernl offices.
Private by default
Self-hosted. Your tasks, email and finances never touch someone else’s cloud. The API always requires a token: there is no unauthenticated mode.
Graph memory
An optional Neo4j + GDS brain links every module for cross-module relationship analysis. Without it, Kernl degrades gracefully.
Any LLM, any client
Claude, OpenAI, Grok, or a local model via LM Studio. Pluggable sandbox drivers and channels: WhatsApp, Telegram, Slack, email, Mattermost and webchat.
Endlessly extensible
Everything is a module. An extension is a manifest plus a backend module, agents, skills, a theme, a channel or a sandbox driver, in a portable .kernl package.
Projects with approval
An office can serve several products. In a project run, anything published or sent is held as a draft in an outbox: nothing goes out until you approve it.
Voice
Talk to the Chief through a microphone and hear the reply. The default engines are local and free: whisper.cpp to listen and Piper to speak.
Mail and WhatsApp
Connect an IMAP/SMTP mailbox from the interface and read messages in the mail view. WhatsApp links with a QR code and the session survives updates.
Positioning
Kernl vs. the usual options
| Cloud assistants | A single MCP server | Kernl | |
|---|---|---|---|
| Data stays on your machine | No | Yes | Yes |
| Covers your real life | Partial | One thing | 71 modules |
| Agents that act, not just chat | Limited | No | Yes |
| Works with any LLM and client | No | Yes | Yes |
| Yours to extend and fork | No | Partial | Apache-2.0 |
Plans
Open-core, no surprises
Core
FreeApache-2.0
- 71 bundled modules
- 103 tools installed, 290+ enabled
- Agents, offices and the 3D view
- MCP over stdio and HTTP
- Community support
All-Access
$12/mo · $10 yearly
- Everything in Core
- Every paid extension unlocked
- Accounting Office and DevOps Office
- MoneyIQ, Web Intel, Habits, Sleep
- New releases as they ship
Install & support
$199from · $49/mo retainer
- We install it on your box
- Modules and channels configured
- Your agent offices set up
- Migration of existing data
- Support with a response agreement
Paid extensions can also be bought one at a time, from $9 to $79, owned forever.
Common questions
What people ask before installing
Does my data ever leave my machine?
Only if you point Kernl at a cloud LLM — and even then, only what that request needs. Run a local model through LM Studio or Ollama and nothing leaves the box at all. There is no Kernl account and no telemetry service to sign up for.
What do I need to install it?
For the one-line install, nothing. The package ships its own runtime: a 64-bit Debian/Ubuntu, Fedora/RHEL or macOS machine is the whole list, and Windows has an .msi. Bun, Node, Docker and Neo4j only come into it if you build from source or run the container stack.
Which LLMs and clients does it work with?
Any MCP client reaches the same tools: Claude Desktop, Cursor, OpenAI, Grok, LM Studio, or your own. Kernl speaks MCP over stdio and Streamable HTTP at the same time, so you do not have to choose.
What exactly is open source?
Kernl is open-core. The kernel, the 71 bundled modules, the MCP server, the agents and the dashboard are Apache-2.0 and yours to fork. The paid extensions — Accounting Office, DevOps Office, MoneyIQ and the rest — are commercial and not redistributable.
Can I write my own modules?
Yes. An extension is a folder with a manifest plus any of: a backend module exposing new MCP tools and HTTP routes, agents and offices, agent-scoped skills, a theme, a channel or a sandbox driver. Package it as a portable .kernl file and install it anywhere.
Give your AI a body
One line to install, four steps to first run, and your agents are working. On your machine, under your licence.