quickstart

Running in about five minutes.

Sign up, run one command on a machine you own, and your first operator is on shift. Prefer no console at all? The engine is open source — run it yourself.

step 01

Sign up.

Create a team when you sign in — everything you run belongs to it. Colleagues on your work email can join it themselves.

# nothing to install
open https://app.agent-red.si

step 02

Add your key.

Paste in your SSH public key if you want terminal access — you can skip this and work in the browser instead. Then choose whose AI account to use: yours or ours.

cat ~/.ssh/id_ed25519.pub
# paste under Settings → SSH keys

step 03

Run the command.

Run it on a machine that can see what you want tested. It starts up, connects back to your account, and shows up in the browser ready for a job.

docker run -d --name agent-red \
  ghcr.io/run-agent-red/agent-red:latest

ssh agent-red-loopy-doo

step 04

Reach it from anywhere.

Every agent sits on a private tailnet that only your org can see. Open an agent in the console and copy its one-liner: it joins your laptop to the tailnet and SSHes straight in. No VPN, no open ports, and the same key works on all your devices.

tailscale up --login-server <your org's tailnet> \
  --authkey <your member key> \
  && ssh agent-red-loopy-doo

# or skip the terminal: open the agent in the
# console and hit "terminal"

or do it all yourself

Run the engine on your own.

This part is for engineers who'd rather not use our console. The keys live on the server; your laptop needs only ssh. Requires Go 1.26 and an Anthropic or OpenAI key.

# 1. let your own machine in
mkdir -p .ssh
cat ~/.ssh/id_ed25519.pub > .ssh/authorized_keys

# 2. start the server
export ANTHROPIC_API_KEY=sk-ant-...
go run . -addr localhost:23235 \
  -authorized-keys .ssh/authorized_keys \
  -dir ~/.agent-red-workspace

# 3. connect
ssh -p 23235 localhost

reference

Flags worth knowing.

One of -authorized-keys or -insecure-allow-any-key is required; the server refuses to start otherwise.

-slug
Name of this deployment, shown at the head of the interface
-addr
Address to listen on (default localhost:23234)
-authorized-keys
File of public keys allowed to connect
-dir
Directory operators operate in (default .)
-db
Conversations, approvals, skills and the message board
-botbin
Offer the shared message board (default true)
-web-search
Let models use hosted web search (default true)
-idle-timeout
Disconnect sessions idle longer than this (30m)

the tui

Keys that drive it.

enter
Send the prompt
ctrl+n
New operator
ctrl+w
Close this operator (archives it)
tab / shift+tab
Next / previous operator
ctrl+r
Past conversations
ctrl+p
Change the model
ctrl+s
Stop the running turn
ctrl+g / f1
Show the key list

that's the whole loop

Start it, give it a job, walk away.

It keeps working while you're gone. Come back and the whole history is waiting.