Give your AI agent a Discord account. discli lets agents send messages, react, manage threads, moderate — all from the terminal. Built with permission profiles, audit logging, and rate limiting so agents can't go rogue. Works just as well for humans who want to script and automate Discord without a GUI.
Hey Product Hunt!
I built discli because I was frustrated with how Discord bots work.
I work on PocketPaw (https://github.com/pocketpaw/poc...), a self-hosted AI agent that runs on your machine with support for Discord, Slack, Telegram, and more. When I was working on the Discord channel, I hit a wall. Every Discord bot library forces you into if-else chains. "If message contains X, reply Y." That's not how an AI agent should work. An agent should think and act on its own. Send messages, react, create threads, moderate, without being hardcoded for every scenario.
So I pulled the Discord layer out into its own tool: discli.
It's a CLI that gives any AI agent (or human) full access to Discord from the terminal. Your agent just runs commands. No bot framework, no event handler boilerplate.
What makes it different:
- Works with any AI agent. Claude, GPT, LangChain, or a bash script. If it can run a command, it can use Discord.
- Security built in. Permission profiles (readonly/chat/full), audit logging, rate limiting, and confirmation prompts for destructive actions. Your agent can't accidentally ban your entire server.
- discli serve mode. Persistent bidirectional JSONL connection for building full bots with streaming responses, slash commands, and real-time events.
- Human-friendly too. Manage your Discord server entirely from the terminal. No GUI needed.
Open source, pip install discord-cli-agent, works on macOS, Linux, and Windows.
Would love your feedback. What would you build with it?
this is actually a cool take. i like that you are not trying to make “another discord bot framework” and instead giving agents a cleaner way to interact with discord directly. the permission profiles and audit logs part also makes it feel way more practical, because that’s exactly the kind of thing people would worry about first.
curious, what are people using it for first in real life, simple community ops stuff or full blown agent workflows?
Rohit Kushwaha