kirbus 0.4.1

P2P end-to-end encrypted terminal chat with AI integration

kirbus

A peer-to-peer, end-to-end encrypted terminal chat with built-in AI integration.

• Private by design — messages are encrypted between peers; no server ever sees content
• No accounts — identity is an Ed25519 keypair generated on first run
• Zero-config start — connect to the public registry and pick a server, no URLs needed
• Works anywhere — direct LAN connections or through a relay server for internet use
• Local AI — /ai command talks to a local Ollama instance; AI context is shared between peers
• Retro terminal UI — five built-in themes, channel support, scratch pad

---

Quick start

pip install kirbus
kirbus --handle yourname

That's it. The client connects to the default registry at kirbus.ai, shows you available servers, and you pick one with Tab + Enter.

Requires Python 3.11+. Or run from source with uv:

git clone https://github.com/wehale/kirbus.git
cd kirbus
uv run kirbus --handle yourname

After install, kirbus, kirbus-server, and kirbus-registry are available as commands. If running from the repo with uv instead, prefix with uv run (e.g. uv run kirbus).

---

LAN mode (no server needed)

# Terminal 1 — listen
kirbus --handle alice --listen 9000

# Terminal 2 — connect
kirbus --handle bob --connect localhost:9000

---

Registry

The registry is a public directory of kirbus servers. The default registry is https://ezchat.kirbus.ai. When you run kirbus --handle yourname, it fetches the server list and lets you pick one.

To use a different registry:

kirbus --handle yourname --registry https://custom.example.com

To skip the registry and connect directly to a known server:

kirbus --handle yourname --server http://SERVER_IP:8000

---

Running your own server

One person runs kirbus-server on a machine with a public IP. Everyone else connects through the registry or via --server.

kirbus-server --config server.toml

Open ports 8000 (rendezvous API) and 9001 (TCP relay) in your firewall.

Example server.toml:

[server]
host       = "0.0.0.0"
api_port   = 8000
relay_port = 9001
ttl        = 60
log_level  = "info"

[registry]
url         = "https://ezchat.kirbus.ai"
name        = "my-server"
description = "A public kirbus server"
secret      = "CHANGE_ME"
access      = "open"
public_url  = "http://YOUR_PUBLIC_IP:8000"

[auth]
mode = "open"

Server access modes

Mode	Description
open	Anyone can join
password	Password required on first connect; pubkey saved for future access
allowlist	Only pre-approved pubkeys can connect

For password-protected servers, add to server.toml:

[auth]
mode     = "password"
password = "your-server-password"

Superuser (admin)

Connect from the same machine as the server with the --su flag:

kirbus --handle admin --su

Su users get admin commands: /kick, /ban, /unban, /who.

---

Running your own registry

kirbus-registry --config registry.toml

Example registry.toml:

[registry]
host          = "0.0.0.0"
port          = 8080
heartbeat_ttl = 180
log_level     = "info"

Servers register themselves via heartbeat. The registry is stateless — listings live in memory and rebuild from heartbeats.

---

Encrypted history

Encrypt chat logs at rest with a passphrase:

kirbus --handle yourname --encrypt-history

First run prompts you to set a passphrase. Subsequent runs prompt for the passphrase to unlock history. The passphrase is never stored — without it, the history files are unreadable.

To decrypt history for export:

kirbus --decrypt-history @alice > alice.log
kirbus --decrypt-history '#general' > general.log

To disable encryption and decrypt everything back to plaintext:

kirbus --handle yourname --no-encrypt-history

---

Multiple identities

Each --handle gets its own data directory (~/.kirbus-{handle}/) with its own keypair, peers, and history:

kirbus --handle work-me
kirbus --handle personal-me

Override with KIRBUS_HOME:

KIRBUS_HOME=~/.kirbus-custom kirbus --handle custom

---

Configuration

~/.kirbus-{handle}/config.toml — created automatically on first run.

[ui]
theme  = "ansi_bbs"        # phosphor_green | amber_terminal | c64 | ansi_bbs | paper_white
handle = "you"
encrypt_history = false

[ai]
provider = "openai-compat"
model    = "gemma3:4b"
base_url = "http://localhost:11434/v1"
api_key  = ""

---

Commands

Command	Description
/help	Show all commands
/ai <prompt>	Ask the AI; response shown in conversation
/ai-peer	Forward the last peer message to your local AI
/theme <name>	Switch theme
/themes	List available themes
/accept [peer]	Accept a new or key-changed peer
/block [peer]	Mark a peer as blocked
/unblock [peer]	Remove blocked mark
/servers	Refresh server list from registry
/connect <name>	Connect to a server by name
/disconnect	Leave current server, return to server list
/channel create <name>	Create a channel
/channel join <name>	Join a channel
/channel invite <peer> [channel]	Invite a peer to a channel
/channel leave <name>	Leave a channel
/clear	Clear chat history
/quit	Exit

Su commands (admin only):

Command	Description
/kick <handle>	Disconnect a peer
/ban <handle>	Kick and revoke access
/unban <handle>	Restore access
/who	List connected peers with details

Keyboard shortcuts:

Key	Action
Tab	Focus peer list
↑ / ↓	Navigate peers or command history
Enter	Select peer / send message
Esc	Return focus to input
PgUp / PgDn	Scroll chat
Mouse wheel	Scroll chat

---

AI integration

kirbus uses Ollama for local AI.

# Install Ollama, then pull a model
ollama pull gemma3:4b

# Ask the AI in any conversation
/ai what's the capital of France?

AI context is shared between peers — if alice asks a question and bob follows up, the AI sees the full conversation history.

/ai-peer grabs the last message from your peer and forwards it to your AI automatically.

Each person's /ai runs against their own local model. For cloud AI, point base_url at any OpenAI-compatible endpoint and set api_key.

WSL2 note: if Ollama is running on Windows, set OLLAMA_HOST=0.0.0.0 before starting it and point base_url at the Windows host IP (find it with ip route | grep default | awk '{print $3}').

---

Verify message signatures

Every message is signed with the sender's Ed25519 key and stored in ~/.kirbus-{handle}/history/.

kirbus --verify-log @alice
kirbus --verify-log '#general'
kirbus --verify-log scratch

---

Test mode

Try the UI without any network setup:

kirbus --test

Simulated peers come online, join channels, and respond to messages.

---

Deploy to AWS

A CDK stack is included in deploy/ to provision an EC2 instance with the registry and a lobby server:

cd deploy
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cdk deploy -c key_name=your-ssh-key

The stack creates an EC2 instance, Elastic IP, security group, nginx reverse proxy, and Let's Encrypt TLS. See deploy/kirbus_stack.py for details.

---

Security model

• Identity — Ed25519 keypair, generated locally, never leaves your machine
• Key exchange — X25519 ECDH ephemeral keys per session
• Encryption — AES-256-GCM with HKDF-derived keys
• Message signing — every message signed by sender's Ed25519 key
• History encryption — optional AES-256-GCM at rest with scrypt-derived key
• Server auth — password gate + pubkey allowlist for access control
• Relay — the relay server pipes opaque ciphertext; it cannot read messages
• Rendezvous — registrations are signed; the server stores handle → IP:port for 60 seconds then discards
• Registry — stateless directory; never touches chat traffic

Self-host the server and registry for full sovereignty.