AI assistants often look like complete products, but in practice many are still just a window with a text box. They can answer, sometimes call a tool, but once you want them to help with real work, something more important is missing: an environment where the agent runs, remembers context, uses tools and reaches you wherever you are.
That is why Elowen exists. A self-hosted personal AI agent you can talk to and let act.
An agent, not just chat
The key difference is that Elowen is not designed as a prompt page. The agent has its own runtime, tools, memory, plugins and several interfaces. You can use it from the CLI, the web dock, Discord or WhatsApp, but it remains one instance.
That matters. If every channel has a different context and different capabilities, the user ends up switching manually again. Elowen pushes in the opposite direction: the same agent, the same tools, the same memory.
Why self-hosted
With AI agents, the model is not the only important part. The important parts are data, permissions and the tools you let the agent use. If an agent can read projects, edit files, run commands or communicate through your accounts, control matters.
Self-hosted means Elowen runs on your own machine or server. It uses your models, your keys and your plugins. It is not locked inside one SaaS service and its capabilities can grow around the actual workflow.
The CLI is home
Elowen includes a full CLI interface. Not as an afterthought, but as the main working surface. In the terminal you can see the live transcript, tool calls, file diffs, task checklist, telemetry panel and modes such as plan mode.
That is important for technical work. When an agent edits code or runs commands, the final answer is not enough. You need to know what it is doing, which tool it is calling and when it is about to change something sensitive.
Web, Discord and WhatsApp
Not everything happens in the terminal. Sometimes a quick Discord message is faster. Sometimes a web dashboard with tasks, sessions, timeline, kanban and live terminals makes more sense. And for some operational scenarios, WhatsApp is the right interface.
Elowen does not treat those channels as separate bots. They are different inputs into the same agent. That makes it possible to use a simple chat for quick work and a richer control surface for larger tasks.
Plugins instead of a hardcoded core
One of the most important parts of Elowen is the plugin system. Capabilities are not baked permanently into the core. Discord, WhatsApp, file tools, terminal tools, the MCP bridge, sub-agents, cron, skills and the security scanner are separate modules.
That keeps the base small while allowing broad capabilities. The agent does not need to have every power all the time. It has what the current installation enables.
Memory, tasks and automation
Elowen can store durable information, recall it by relevance and use it in later conversations. It can keep a todo checklist for multi-step work, plan, delegate parts of a task to sub-agents and run scheduled or one-shot prompts.
That turns it into more than an answering tool. It becomes a personal operating layer over repeated work: project checks, website maintenance, writing articles, small operations, internal routines or customer communication.
Where it is going
Elowen is aimed at a simple idea: an AI agent should be your system, not someone else’s box. It should run where you want it, have the capabilities you give it and be reachable through interfaces you actually use.
That feels like an important direction for practical AI. Not more chats. Not another panel pretending to be a revolution. An agent that can work with context, tools and responsibility inside one controllable environment.
The repository is available on GitHub: github.com/dragocz95/elowen.