Learns from every task
SARTHI creates reusable skills after complex work, improves them during use, and nudges itself to persist what it learned. Your agent on day 30 is smarter than on day 1.
Watch it learn
A real session: SARTHI does the work, notices the pattern, automates it — and keeps the skill forever.
sarthi — terminal
❯
One agent. A closed learning loop.
Remembers who you are
Persistent memory, session search, and a deepening model of your preferences across conversations. You never re-explain yourself.
Lives where you do
Terminal, desktop app, Telegram, Discord, Slack, WhatsApp, Signal — one agent, one memory, every platform. Send a voice memo; it just works.
Any model, no lock-in
OpenRouter, Nous Portal, OpenAI, NVIDIA NIM, Hugging Face, or your own endpoint. Switch with one command: sarthi model.
Works while you sleep
A built-in cron scheduler runs daily reports, nightly backups, and weekly audits — written in plain language, delivered to any platform.
Runs anywhere
Your laptop, a $5 VPS, Docker, SSH, or serverless clouds that hibernate when idle. Your agent isn't chained to one machine.
Running in three commands
Step 1 — Install in one line
$ curl -fsSL https://sarthi-agent.vercel.app/install.sh | bashStep 2 — Pick your model & keys
$ sarthi setupStep 3 — Start talking
$ sarthiYour charioteer for the
age of AI.
Sarthi (सारथी) means charioteer — the trusted guide who knows the road and knows you. Free forever, open source, yours.
Get SARTHI free