A Postgres database that stays up.
The easy way to run highly available, multi-node PostgreSQL on your own computers. One small app on two or three of them keeps real PostgreSQL 18 in sync between them, built so your database stays up when one stops.
The idea
Like a pack that keeps in touch.
Coyotes spread out across a valley and still know where the others are. CoyoteDB's nodes work the same way. Watch one turn into the other.
The packCoyotes spread out along a ridge keep in touch by calling to each other across the valley.
CoyoteDBPut a computer where each one stands. Those are your nodes: the main node takes the writes, and the others keep live copies and keep checking on it.
The packOne coyote goes quiet. The rest of the pack notices and carries on.
CoyoteDBIf the main node stops, the others vote and one of them takes over (with 3 or more nodes, or 2 and a witness). The database keeps going.
The packA newcomer is let into the pack.
CoyoteDBA new computer joins with an invite code like coyote-7k4q-2xpd, and gets its own full copy.
How it works
Three steps. No database expert needed.
Start one app
Start the small CoyoteDB app on your first computer. It downloads, checks and runs PostgreSQL 18 for you.
Paste an invite code
Ask the first node for an invite code. Paste it on your next computer. The two find each other and start syncing.
It stays up
Every save is copied to another node. With 3 computers, or 2 plus a witness, another node becomes the main node by itself when one stops, and the database carries on.
# on the first computer
$ coyote up
$ coyote invite
coyote-7k4q-2xpd
# on the second computer
$ coyote join <code>
# on either one
$ coyote status
Why CoyoteDB
One computer is one failure away.
Many apps keep their database on a single computer. CoyoteDB is built to be the easy way to put it on two or three, so it stays up. Why it is built this way
What problem does CoyoteDB solve?
Many apps keep their database on one computer. When that computer dies, the data and the app go with it. Making Postgres highly available usually means putting several separate tools together and having a specialist look after them.
Why does it exist?
So a small business or a solo builder can have a database that stays up without that expertise: install one small app on two or three computers and they look after each other.
Why real PostgreSQL instead of something new?
Because every Postgres tool, library and hosting habit already works with it, and your data sits in a format that has been trusted for decades. CoyoteDB adds the copying and the switch-over around it; it does not replace it.
Why on your own computers?
So the data stays with you, on machines you control, with no cloud database to rent. The computers can be in one room or linked over a private network such as Tailscale or a site-to-site VPN.
Built for
High availability, without a database specialist.
From the app in a shop's back office to a homelab on Tailscale: who CoyoteDB is built for, and the setup we would suggest for each. All use cases
The app in the back office keeps running.
Your orders, bookings, stock or till app runs on one PC in the back office, and if that PC dies, the business stops.
A second computer keeps a full, live copy of the database. With a witness, it takes over by itself if the first one dies, so the database carries on without it.
Suggested setup 2 computers + a witness
More on this use caseReal Postgres you own, on your own machines.
You are building an app, often with an AI coding agent, and you want a real Postgres database without renting one from a cloud.
CoyoteDB runs PostgreSQL 18 on your own computers, so it is hosted where you are and there is no cloud database to pay for each month. Any Postgres library connects to it, and your coding agent can use it like any Postgres database.
Suggested setup Start with 1 computer, grow to 2 + a witness
More on this use casePostgres across your own boxes.
The services you run at home all depend on one Postgres, on one mini PC that will fail one day.
Spread Postgres across your own machines over Tailscale or your LAN. When a mini PC dies, the others hold full copies and carry on by themselves.
Suggested setup 3 computers, or 2 + a witness on a Raspberry Pi
More on this use caseStart small, add computers later.
You need a database today, but you do not want to plan the whole setup up front.
Start on one machine. When it matters more, add computers with an invite code. Your app connects the same way before and after.
Suggested setup Start with 1, add computers with an invite code
More on this use caseA dependable database at your client’s site.
You build software for clients who need it to keep running, but nobody there is a database specialist.
Put one app on the client’s computers and join them with invite codes. If one computer stops, the others take over by themselves (with 3 computers, or 2 plus a witness), without someone on call to fix it.
Suggested setup 2 computers + a witness
More on this use caseA copy in each location.
You have two sites, and you want the data to survive if one of them loses power or its internet line.
Link computers in each site with a site-to-site VPN, so each one holds a full copy, and put the witness in a third place or on a small cloud server. It is built for links like a VPN between a few sites, not for spreading a database across the world.
Suggested setup 2 computers + a witness in a third place
More on this use caseSetups
Pick a setup. Then unplug a computer.
Each computer running CoyoteDB is a node. Click one to see what happens when it goes quiet.
Click a computer to unplug it. Click again to plug it back in.
Database is up
Recommended setups
Most people: two computers and a witness.
A witness is any small always-on device, like a Raspberry Pi. It holds no data and only votes, so switch-over happens on its own. When the data really matters, use three computers.
What's inside
Small app. Real Postgres.
Everything you need to run it yourself, and nothing you have to learn first.
Real PostgreSQL 18
Real Postgres runs inside. Any Postgres tool or library works with it.
One small app, self-hosted
One program on your own machines. It downloads, checks and runs Postgres for you, and your data stays there.
Invite codes
Ask one computer for a code, paste it on the next, and they join over one encrypted port.
Connect to any node
Every node will pass you on to the current main node, so you never need to know which one that is. Not available yet
Web dashboard
A Create / Join setup wizard and a health view, in your browser. Not available yet
macOS and Linux
Tested on both. Windows has not been tested yet. Data nodes in one cluster share one system type.
One limit to know: Today, the computers that hold data in one cluster must share one system type: all Windows, all Linux, or all Mac with the same chip family. Support for mixing Windows, Linux and Mac data computers is being built and tested now. Until it has been proven safe, CoyoteDB refuses a mixed join rather than risk your data. The witness and the apps that connect can already run on any system.
Networks
Works on the network you already have.
Your computers just need to reach each other. Nodes talk to each other over one encrypted port.
- Tailscale
- Twingate
- UniFi / site-to-site VPN
- Plain LAN
Coding agents
Ready for your coding agent.
Your coding agent can use CoyoteDB like any Postgres database: coyote conninfo prints the connection details, and a short note tells the agent how to use them.
# print the connection details
$ coyote conninfo