Every big AI company now wants to be your personal assistant, and every one of them keeps your memory on its servers. Saathi is the other way round: an open-source agent that runs on a server you control, watches the sources you care about, and reports only to you. The name is Hindi for companion.
- Runs on
- Your server
- Monthly cost
- $0
- Licence
- MIT
- Talks to you in
- Discord
- Built on
- Hermes Agent
What mine does before I wake up.
Each job posts to its own private channel. Every news claim carries its source link, and anything it cannot cite is dropped.
- 06:30
AI news, the whole field
Models, labs, funding, policy and open source, from lab blogs, Hacker News, Reddit and research feeds.
- 07:00
Trending GitHub repos
Ranked by stars actually gained, measured from its own daily snapshots rather than a headline number.
- 07:30
One deep topic a day
RAG on Monday, agents on Tuesday, continuous training on Wednesday, audio and video on Thursday, evals on Friday.
- live
Markets and alerts
Pre-market and after-close briefs, economic-calendar reminders, and a ping on my phone when something big moves. Research, not advice.
- 08:30
Content ideas and drafts
Ideas for LinkedIn, reels and YouTube, and drafts I review before anything is posted. It never posts on its own.
What stays with you, and what does not.
The honest version, because the comments will ask.
On infrastructure you control
- Its memory, skills and schedules
- Your config, your sources and every output
- Your keys, in a private file only the agent user can read
What still leaves the server
- Each prompt goes to the model provider you choose. Free models can log prompts
- Run a local model on your own machine if you want nothing to leave at all
- Free models are weaker than paid frontier ones. That is the trade for $0
Build your own.
Saathi is open source. It asks what you want your agent to do, finds the sources, shows you a plan, and changes nothing until you say yes. Later you can change it by just asking: add a ticker, drop a source, make the drafts shorter.
The repository has the step-by-step setup on a free cloud server, a guide to choosing other agents, models and hosts, and every problem I hit on the way, with the fix.