One command. It works on a machine with nothing on it — no Python, no pipx, no
git, no clone — and it never asks you anything, so it is safe to run from a
script.
bash
curl-fsSLhttps://lesysbot.github.io/install.sh | sh
powershell
# Windows
irm https://lesysbot.github.io/install.ps1 | iex
Here is everything it does, in order:
Finds a Python
3.11 or newer. If there isn't one it fetches a private copy with uv — no system packages, no root.
Builds an isolated environment
~/.local/share/lesysbot/venv, so LeSysBot's dependencies can't collide with anything else you have installed.
Puts lesysbot on your PATH
A link in ~/.local/bin, added to your shell startup files if it isn't there already (--no-modify-path to skip).
Gets a model ready
Installs Ollama and pulls qwen3.5:4b — small, runs on almost anything, reliable at picking the right tool.
Configures everything
Runs lesysbot setup --yes: writes ~/.lesysbot/config.yaml, installs the bundled tools and dashboards, brings up the Grafana stack, and starts the background service.
When it finishes you have a working bot. Skip to Say hello.
Read it first? Sensible — you should with any curl | sh.
Every flag also has a LESYSBOT_* environment variable, because a bare
curl … | sh can't easily take arguments — LESYSBOT_SKIP_DASHBOARD=1,
LESYSBOT_MODEL=…, and so on.
Nowhere that LeSysBot can avoid, and it never asks for one. On macOS and
Windows Ollama installs without elevation, so the installer just does it. On
Linux Ollama's own installer needs root — so unless you're already root (or
have passwordless sudo), the installer skips it and prints the two lines
for you to run:
The tools, the dashboards and the Prometheus/Grafana stack all ship inside the
package, so this works offline and lesysbot update can refresh them later.
From a git checkout — for working on LeSysBot, or to run an unreleased
version. Run from a checkout, the installer installs that checkout:
lesysbot setup --yes is what the installer runs. It takes every default
without prompting and reads overrides from the environment, so a complete
Telegram install is one command with no terminal involved:
bash
LESYSBOT_SETUP_PROVIDER=telegram \
LESYSBOT_SETUP_TELEGRAM_TOKEN=123456:ABC… \
LESYSBOT_SETUP_TELEGRAM_ALLOWED_IDS=123456789 \
curl-fsSLhttps://lesysbot.github.io/install.sh | sh
Variable
Default
LESYSBOT_SETUP_LLM
ollama — or openai, vllm, custom
LESYSBOT_SETUP_BASE_URL
per backend (http://localhost:11434/v1 for Ollama)
A missing required value aborts and names the variable to set, rather than
leaving a half-configured bot that anyone can message.
Re-running lesysbot setup --yeskeeps your existing config.yaml and
refreshes everything else (tools, dashboards, service) — that's the upgrade
path. --reconfigure replaces the config instead.
Run lesysbot setup with no flags anytime to change your answers through
the interactive wizard. No reinstall needed.
Every question the interactive wizard asks, explained
Each prompt shows a default in [brackets] — press Enter to take it. Menus
respond to ↑/↓ + Enter, or just press the option's number. From the second
question onward every menu has a ← Back entry, reachable with the ← key
or Esc. Esc at a typed prompt (URL, model, token) abandons it and returns to
the menu, so you can never get stuck.
"~/.lesysbot/config.yaml already exists — overwrite?" [y/N]
Only appears if you've installed before. n keeps your current settings and
skips to the startup question; y starts fresh. This protects a config you may
have hand-edited.
"LLM Backend" [1]
1) Ollama — local, recommended (no API key needed)
2) OpenAI — cloud API
3) vLLM — self-hosted OpenAI-compatible server
4) Custom — any OpenAI-compatible endpoint
All four speak the same protocol, so this only picks a URL and a key.
Ollama — the wizard runs ollama list and shows the models you already
have as a menu. The last entry lets you type any model name and pulls it on
the spot. No models yet? It offers to download one (default qwen3.5:4b). No
Ollama CLI on PATH? It asks for a name to use once Ollama is available.
OpenAI — asks for a model (gpt-4o) and your API key. The key is stored
in config.yaml, so keep that file private.
vLLM — asks for the base URL (http://localhost:8000/v1) and the model id
your server serves.
Custom — anything else OpenAI-compatible (LM Studio, llama.cpp server, a
proxy). Asks for the full URL including /v1, a model id, and an optional key.
"How to reach LeSysBot" [1]
1) Terminal only (default)
2) Telegram
3) Discord
The terminal always works regardless of this choice — lesysbot chat
opens a chat whatever you pick here. Telegram and Discord are extra remote
channels that run in the background.
Telegram — asks for a bot token from
@BotFather and the numeric user IDs allowed to use
the bot (find yours via @userinfobot). At least one
ID is required; the wizard re-asks until you give one, because an empty list
means anyone who finds your bot can run your tools.
Discord — asks for a bot token and the numeric user IDs allowed to use
the bot, the same way. Create the bot at
discord.com/developers/applications
and turn on the Message Content intent — without it the bot can't read your
messages. Full setup: Telegram & Discord §3.
"Service" — start now, or also at every reboot? [1]
Option 1 starts it now and again on every boot; option 2 starts it now only.
Every setup gets the service, whichever channel you picked: it serves the
control panel (settings, tools, health) and, with Telegram or
Discord, receives your messages. A terminal chat is still yours to start — the
service never opens one.
Summary — apply, change, or quit
A final menu showing your model, provider, startup behaviour, and where the
config will go. Apply writes everything. The Change … entries jump back
into any step with your previous answers as the defaults. Quit exits without
writing anything.
The Grafana dashboard — set up during install
After it writes the config, setup also seeds the
dashboard stack into ~/.lesysbot/dashboard and
gets you to a Grafana dashboard at http://localhost:3000. It first asks how
you want it set up (see per-OS below), then the Grafana username and password
LeSysBot should use to reach it (defaults admin / admin; the password is
hidden as you type). Those are saved to ~/.lesysbot/grafana.env, which LeSysBot
loads at startup — so the share_dashboard tool and the status screen
authenticate automatically. This is a standard part of LeSysBot, not an opt-in —
and always no-sudo, never fatal to the install:
Linux — Docker is the path. If Docker is already running, setup asks
whether to auto-start the bundled Prometheus + Grafana stack now, or set it
up manually later. The bundled Grafana boots with the username/password you
entered. If Docker isn't ready, it prints the exact steps to get it going
(install Docker Engine, start the daemon, or join the docker group) — or run
Grafana natively instead.
macOS — Homebrew is the path, and it doesn't require Docker Desktop.
Setup asks whether to install it now, then brew installs Grafana, Prometheus
and node_exporter, wires the datasource and dashboard up, sets Grafana's
admin password to the one you entered, and runs all three under brew services so they survive a reboot. Nothing else to do — open
http://localhost:3000. If Homebrew isn't installed, setup says so and falls
back to the manual instructions below.
It asks one extra question here: whether to install a small helper for
CPU/GPU die temperature. macOS publishes that only through a private
framework or root-only powermetrics, and LeSysBot never uses sudo, so those
two tiles need macmon (Apple Silicon) or smctemp (either chip). The answer
defaults to no, only the tool that can work on your Mac is offered, and a
failed install never fails the setup — every other panel works without it, and
you can add one at any time. Answer up front, or skip the prompt entirely on an
unattended install, with LESYSBOT_TEMP_HELPER=macmon|smctemp|none.
Windows — setup doesn't require Docker Desktop. It warns and walks you
through a native Grafana install from
grafana.com/grafana/download: install
it, open http://localhost:3000, and set Grafana's admin login to the
username/password you entered so LeSysBot connects (it detects Grafana on
port 3000 automatically; set LESYSBOT_GRAFANA_URL only if it runs elsewhere).
If you do have Docker running, it also points out the one-command bundled
stack as a shortcut.
Set LESYSBOT_SKIP_DASHBOARD=1 before running setup to skip this step entirely
(e.g. an unattended install that shouldn't pull images or prompt).
Re-running setup is how the stack gets updated. From a checkout, it refreshes
the shipped scripts, dashboards and compose files in ~/.lesysbot/dashboard
whenever they've changed upstream, while leaving the two things you own alone
forever: .env (ports, Grafana login) and prometheus/ (any scrape targets you
added). Re-run your OS's start script afterwards so the dashboard is rebuilt with
the new code.
Install by hand instead (no wizard)
Three steps: install the package, write a config, run it.
Install the package
bash
pipinstall".[all]"# terminal + Telegram + Discordpipinstall . # terminal and tools only — smallerpipinstall-e".[dev]"# development install (adds pytest + ruff)
Extra
Adds
Needed for
telegram
python-telegram-bot
--provider telegram
discord
discord.py
--provider discord
all
both
what the install scripts use
Pick a provider you didn't install and LeSysBot tells you which extra to add
rather than crashing.
Write a config
bash
cpconfig/default.yamlconfig.yaml
The parts that matter:
yaml
messaging:
provider: cli # cli | telegram | discordllm:
base_url: "http://localhost:11434/v1"# Ollama defaultmodel: "qwen3.5:4b"# a model you've pulled (ollama list)api_key: "ollama"# any non-empty string for Ollama/vLLM
For OpenAI, change those three lines to
https://api.openai.com/v1, gpt-4o, and your real key. Everything else is in
Settings.
Run it
bash
lesysbotrun# control panel + bot, using ./config.yamllesysbotchat-v# force a terminal chat, verboselesysbot-c /path/to/config.yaml# status for a config somewhere else
If you want it running in the background, set that up yourself:
Run as a service.
You: what is 2 + 2?
Bot: 4.
You: what's the disk usage of /tmp?
Bot: /tmp has 2.3 GB free out of 20 GB — 11% used.
You: /disk_usage path=/tmp
Bot: Path: /tmp Total: 20.0 GB Free: 2.3 GB Used: 11.0%
Those last two do the same thing two different ways. The first asked the model,
which chose a tool. The second called the tool directly — no model involved, so
it works even when Ollama is off.
Type /help to see every tool it currently has.
Running as a Telegram/Discord service already? lesysbot chat still opens a
separate terminal chat alongside it. They don't conflict. (lesysbot chat is
short for lesysbot chat, which also still works.)
A running bot picks them up immediately. More in
Install tools.
Or write one. Create ~/.lesysbot/tools/hello/tool.py:
python
from lesysbot.mcp import tool
@tool(description="Say hello to someone")
asyncdefhello(name: str) -> str:
return f"Hello, {name}! Nice to meet you."
Save it. It's live:
You: /hello name=World
Bot: Hello, World! Nice to meet you.
You: say hello to Alice
Bot: Hello, Alice! Nice to meet you.
Shell commands work too, with no Python logic:
python
from lesysbot.mcp import CLITool
df = CLITool(
name="df",
description="Show disk usage for a filesystem path",
command="df -h {path}",
params={"path": "Filesystem path to check"},
)
The control panel is always on — the background service serves it:
http://127.0.0.1:8700
Edit settings and toggle, install, or remove tools there. It's bound to
127.0.0.1, so it never appears on your network. See
Control panel.
To check on things from a terminal instead, run lesysbot with no arguments: it
prints health and metrics — backend, tools, service, panel, Grafana — and exits.
It works backwards through what the installer did: stops and removes the
background service, stops the Grafana stack (without removing its Docker
volumes, so your stored history survives a reinstall), deletes the lesysbot
command and its environment, and takes its entry back out of your shell startup
files.
It keeps ~/.lesysbot — your config, tools, dashboards and logs — so a
later reinstall finds everything as you left it. Add --purge (-Purge on
Windows) to delete that too.
Installed some other way? Then remove it that way — pipx uninstall lesysbot,
or delete the virtualenv you made — and rm -rf ~/.lesysbot when you're done
with the data.
