Durable work units
Kanban tasks and long-running missions with plans, todo state and evidence you can inspect mid-run.
One agent that can talk, plan, use tools and carry real work across your projects.
Elowen turns a conversation into visible work. It plans the next steps, calls tools, delegates specialists and returns with evidence — while your models, memory and permissions stay under your control.
Ask a question, drop in a repository, or describe an outcome. Elowen keeps the conversation connected to the work: current project, active goal, tools, memory and the decisions already made.
You can stay in the terminal when speed matters, or open the WebUI when you want the whole system in view.
Meet the CLI →
Long tasks do not disappear into a spinner. Elowen frames the goal, builds a plan, updates its todo state, calls tools and records the evidence it needs to finish.
Tasks and missions stay inspectable while they run. You can steer the next step, review a diff, open a commit or stop before a risky action.

Elowen can delegate focused work to sub-agents without fragmenting the mission. A coding specialist can inspect the repository, a reasoning model can challenge the plan, and another agent can verify the result.
Each specialist may use a different configured model and a scoped set of tools. Their findings return to one goal, one context and one accountable result.
How autonomy works →Elowen keeps the facts, decisions, preferences and project context worth carrying forward. Memory is visible and editable — not a hidden profile assembled somewhere else.
Search it, group it, connect related facts, or remove what no longer belongs. The agent remembers because you chose to keep it.
Explore memory and chat →
Projects define where Elowen works. Users and roles define who can ask. Tool rules and approval gates define what may happen without confirmation. Provider credentials stay on the daemon.
The same boundaries follow the agent through the CLI, WebUI and connected channels. There is one runtime and one policy — not a different agent behind every screen.
Read the security model →
Use the CLI for focus, the WebUI for context, and plugins for the channels and services your team already lives in.
CLIFast conversation, goals, todo state and live tool output.
WebUITasks, projects, memory, sessions, editor, terminal and system control.
ChannelsDiscord and WhatsApp connect back to the same agent runtime.
PluginsAdd skills and integrations without forking the brain.
Kanban tasks and long-running missions with plans, todo state and evidence you can inspect mid-run.
Delegate focused work to specialists on different models — findings return to one mission context.
From ask-first to autopilot, with an overseer gate deciding what may proceed without you.
Decisions, preferences and project context — searchable, editable and deletable, never a hidden profile.
Registered repositories, inspectable diffs, PR workflow and handoff notes between agent runs.
Users, roles, per-user tools and models, granular tool permissions and approval gates for risky actions.
Discord and WhatsApp bridges plus scoped plugin capabilities — one runtime behind every surface.
A chat TUI for speed and a full web workspace: kanban, sessions, editor, terminal and settings.

Not a demo, not a toy dataset. These run on Elowen every day.
Patricie runs a hair salon. Clients message her Elowen on WhatsApp — it offers open slots, confirms the booking with an SMS code and sends a reminder the day before. Payments and expenses land in the books on their own. She reads the morning summary and just cuts hair.
The documentation you're reading was written, restructured and deployed by Elowen itself — 28 pages, live on a Sunday evening. The orbital menu up top? Its own idea, ported from its own WebUI.
At J+P Autodíly, a supplier's delivery note gets photographed at the counter. Elowen reads the barcodes, matches the parts against the catalog and stocks them in. What used to take an evening of typing takes minutes.
Most agent harnesses assume one user, one machine, one API key. Elowen assumes a life — and maybe a team.
One command installs everything — Node.js, the daemon, tmux, the systemd services and the first admin — and leaves a running WebUI. It works on Linux, or on Windows through WSL2. Prefer to do it by hand? A guided npm route and a Docker image are in the docs.
$ curl -fsSL https://raw.githubusercontent.com/dragocz95/elowen/main/install.sh | bash
# Linux — installs Node, tmux, systemd, admin
● Elowen is live — http://localhost:4500
$ elowen # opens the chat TUI $ npm i -g elowen
$ elowen # guided setup wizard, then the chat TUI
● WebUI at http://localhost:4500 $ git clone https://github.com/dragocz95/elowen.git && cd elowen
$ docker build -t elowen .
$ docker run -d -p 4400:4400 -v elowen-data:/app/data elowen
# one SQLite file on the volume — upgrades are a container swap # inside WSL2 (Ubuntu is installed for you if needed)
$ curl -fsSL https://raw.githubusercontent.com/dragocz95/elowen/main/install.sh | bash
● WebUI reachable from Windows at http://localhost:4500 On your machine. Elowen is a self-hosted daemon with a single SQLite database file — conversations, memory, tasks and settings never leave your infrastructure. Provider credentials stay on the daemon and are never exposed to clients.
Any provider you configure — Claude, GPT, Gemini and others. Each sub-agent can run a different model, and administrators can restrict which models each user is allowed to use.
Yes, to the degree you allow. Autonomy levels range from ask-first to autopilot, an overseer gate decides which steps may proceed without confirmation, and approval gates protect risky actions.
Elowen ships with users and roles (RBAC), per-user tools and models, project assignment and visibility rules. Everyone talks to the same runtime under the same policy.
The recommended setup is a systemd service behind a reverse proxy — sudo elowen install sets it up, and the one-line installer runs it for you. A Dockerfile is available if you prefer containers.
Yes, through WSL2. Elowen runs its agents in tmux and its services under systemd, both Linux-only, so the Windows installer enables WSL2, installs Ubuntu if needed, and runs the exact same Linux install inside it. The WebUI is then reachable from Windows at http://localhost:4500.
Yes — MIT licensed, developed in the open at github.com/dragocz95/elowen.