by vgent

Discord channels for your Claude Code agents.

Aquila gives every agent its own Discord bot, its own channel, and its own working directory — then runs them. Message #backend and the agent working in that repo answers. Two commands from nothing to a running fleet.

Quick start View on GitHub

aquila — zsh

$ aquila init backend=~/src/api frontend=~/src/web Create 2 application(s) at https://discord.com/developers/applications (New Application → name it anything → Bot → Reset Token → copy) backend token › •••••••••••••••••••••••• ✓ app "Orders API" frontend token › •••••••••••••••••••••••• ✓ app "Storefront" ✓ backend: message content intent on ✓ frontend: message content intent on ✓ found server "acme-dev" ✓ owner 1021487254104973352 Install the remaining 1 bot — your server is pre-selected, so each is a single Authorize click: frontend https://discord.com/oauth2/authorize?… ✓ frontend: joined ✓ backend: shows as "backend" in the server ✓ frontend: shows as "frontend" in the server ✓ #backend → /home/you/src/api ✓ #frontend → /home/you/src/web ✓ discord channel plugin installed 2 agent(s) ready. Start them with: aquila up $ aquila up · backend starting (pid 91820) · frontend starting (pid 91821) ✓ backend connected → #backend ✓ frontend connected → #frontend

✦

One bot per agent

Each agent is a real Discord bot, so @backend autocompletes, presence and typing indicators are per-agent, and replies thread natively. Webhook impersonation gets you none of that.

◈

Isolation Discord enforces

Every channel is scoped so only its own agent can read it — cross-channel reads return 403. Not a rule in our routing code that a bug could leak through; a rule Discord applies.

◐

No snowflake hunting

The server id comes from the first bot's own guild list, and your user id from that server's owner. No Developer Mode, no right-click Copy ID, no pairing codes.

Quick start

Create a Discord server

In the Discord client: + → Create My Own. Three clicks, once — not per agent. Bots can't create servers, so this one is yours.
Create one application per agent

At the Developer Portal: New Application → name it anything → Bot → Reset Token → copy. About 40 seconds each. Name it whatever you like; Aquila sets a server nickname so each bot displays as its agent name.
Provision
```
aquila init backend=~/src/api frontend=~/src/web
```
Paste each token when prompted. Aquila enables the Message Content intent, hands you a one-click install link per bot, creates a scoped channel each, and writes access policy and tool pre-approval.
Run them
```
aquila up
```
Returns in about a second. Each agent runs as a detached session with its own pty and its own bot token, and outlives your shell. Message #backend and it answers.

Commands

Command	What it does
`aquila init <agent…>`	Provision bots and channels. Accepts `name` or `name=path`.
`aquila add <agent> [path]`	Add one more agent to the server you already have.
`aquila up [agent…]`	Start agents. Returns immediately; verifies each gateway actually connected.
`aquila down [agent…]`	Stop agents. Confirms the process group actually exited.
`aquila status`	Agents, channels, and session state. Flags any config/reality drift.
`aquila move <agent> <path>`	Change working directory — carries settings, checks trust, restarts.
`aquila set <agent> k=v`	Change a setting, warning when a restart is needed to apply it.

Useful flags

Flag	Effect
`--web`	Paste tokens into a browser form instead of the terminal.
`--trust`	Accept Claude Code's folder-trust prompt for each agent's directory.
`--adopt`	Take over an existing channel that already has the agent's name.
`--rename`	Also change the bot's global username (rate-limited; rarely wanted).

A shared room

Agents can join a common channel alongside their private one. With requireMention on, each responds only when addressed — and because these are real bot users, @backend autocompletes in Discord's composer.

They can read each other there: channel history comes back unfiltered, including other agents' replies. They can't wake each other, so you stay the scheduler — ask one agent something, then ask another to build on it.

What Aquila can't do

Two things, both enforced by Discord. We'd rather say so here than have you discover them halfway through setup.

It can't create the Discord application

There is no public endpoint. The Developer Portal drives private routes with a user token, and automating that means driving a user account — self-botting under Discord's terms, with real account-termination risk. Aquila doesn't do it, and neither should anything else you install. Budget about 40 seconds per agent.

It can't create the server

Discord closed POST /guilds to applications on 2025-07-15 — bots now get Bots cannot use this endpoint, and guilds that bots did own were transferred to real users. So you create the server. Three clicks, once, not per agent.

Everything between those is automatic: intent flags, install links with your server pre-selected, scoped channels, access policy, tool pre-approval, and the sessions themselves.

Built on the official channel plugin

Aquila sits on top of Anthropic's discord channel plugin rather than replacing it — writing its config, and launching sessions with --channels. Per-agent isolation comes from giving each one its own state directory.

View on GitHub