How I Set Up 4 AI Agents on One Discord Server

TL;DR

Running multiple AI agents on one Discord server requires separate bot tokens, dedicated channels, allowlist routing, and isolated workspaces. Skip these and you'll burn 4x the API costs while your agents argue with each other.

Key Takeaways

1. 1.Each agent needs its own Discord bot token — sharing tokens causes agents to respond as each other and cross contexts.
2. 2.Channel isolation is the single biggest cost saver — restricting agents to their own channels cut API costs by ~75%.
3. 3.Use allowlist group policy with requireMention on specialist agents so they only activate when tagged, not on every message.
4. 4.Keep workspaces separate but symlink shared files — isolated memory prevents one agent from referencing another agent's conversations.

i just set up 4 AI agents on one Discord server. each with its own bot, its own personality, its own dedicated channels.

took me 6 hours. would've taken 2 if i knew what i know now.

here's everything i learned, including the mistakes that cost me real money, so you can skip the pain.

the setup

one machine running an AI gateway. four agents, each with a different job:

• one handles general operations and coordination
• one does AI and tech research
• one tracks health and biometrics data
• one covers legal and compliance

each agent gets its own Discord bot, its own workspace, its own memory. they can share files when needed but think completely independently of each other.

sounds simple. it wasn't.

step 1: create separate bots

go to discord.com/developers/applications and create a separate application for each agent.

yes, each one needs its own bot. do not try to run multiple agents through one bot token. when we tried this, agents were responding as each other. messages got crossed. one agent would answer a health question with trading advice because it was reading from the wrong context. total chaos.

for each application:

• click into the Bot section
• generate and save the token somewhere secure
• enable the Message Content Intent (required to read message text)
• go to OAuth2, generate an invite link with bot scope
• give it Send Messages and Read Message History permissions
• invite each bot to your server

now you have 4 bots sitting in your server doing nothing. time to wire them up.

step 2: channel isolation

this is where most people screw up.

create dedicated channels for each agent. your research agent gets a research channel. your health agent gets a health channel. your legal agent gets a legal channel. your main agent gets general.

if your agents can see every channel, every single message in every channel triggers every agent. that means 4x the API calls, 4x the token burn, 4x the cost.

we learned this the hard way. within the first hour of going live, all four agents were reading and responding to the same messages across all channels. our API usage spiked immediately.

here's the math: 50 messages a day across all channels. if all 4 agents process all 50 messages, that's 200 API calls per day. each call sends the full conversation context — 50K to 100K tokens. you're burning tens of millions of tokens per day on messages that aren't even meant for that agent.

the fix: restricting each agent to only the channels it needs. that single config change cut our costs by roughly 75%.

step 3: configure routing and permissions

this is the actual architecture that makes it work. in your gateway config, each agent needs two things.

first, an account entry with the bot token and a restrictive group policy. set it to allowlist mode, not open. open means the agent listens to everything. allowlist means it only responds in channels you explicitly approve.

second, a binding that maps agent to channel. this tells the system: when a message arrives in channel X, route it to agent Y using bot token Z. no binding means the agent ignores the channel completely.

one more critical setting: requireMention for your specialist agents. set this to true. this means the agent only activates when someone tags it, not on every message that flows through. your main agent can run without requireMention in its home channel, but your specialists should absolutely have it on.

without requireMention, our research agent was jumping into casual conversations in shared channels and writing 500-word responses to messages like "lol." not great.

step 4: separate workspaces, shared files

each agent needs its own workspace directory with its own memory files, identity docs, and conversation history.

but they often need to read from common files. your research agent might need to check what tasks are active. your health agent might need access to shared reference docs.

the solution: symlink shared directories into each agent's workspace. your research agent can read the same task files your main agent writes. but their conversation histories and memory stay completely separate.

this matters because if one agent's conversation context bleeds into another agent's, you get unpredictable behavior. agent A starts referencing conversations it never had. agent B gives advice based on agent A's personality file. keep the brains isolated, share only the data.

the mistakes so you don't repeat them

mistake 1: one bot token for all agents. agents were responding as each other. a question in the health channel got answered by the research agent because they were all sharing one identity on Discord.

mistake 2: open group policy. every agent was listening to every channel. token usage went through the roof within minutes. we were burning API credits on agents processing messages that had nothing to do with their job.

mistake 3: no requireMention on specialists. agents were responding to every message in every channel they could see. casual "hey" messages were getting 3-paragraph AI responses from agents that had no business being in that conversation.

mistake 4: shared workspaces. one agent's memory files were getting read by another agent, creating confused and contradictory responses. agent A would reference something agent B discussed yesterday because they were reading from the same memory directory.

every single one of these is a configuration fix, not a code change. that's the good news. the bad news is figuring this out by trial and error costs real money in API usage.

the result

after fixing everything: 4 agents, each in its own lane.

the research agent posts AI news summaries to its channel automatically. the health agent tracks biometrics data and answers questions in its channel. the legal agent handles compliance queries. the main agent coordinates everything and lives in the general channel.

they don't step on each other. they don't waste tokens reading messages meant for someone else. and when you tag a specific agent in a shared space, only that one responds.

the checklist

• separate bot token per agent — do not share tokens
• dedicated channels per agent — do not let them see everything
• allowlist group policy with requireMention on specialists — do not let them respond to everything
• separate workspaces with symlinked shared files — do not let memories cross
• explicit bindings mapping each agent to its channels — do not rely on defaults

the architecture matters more than the AI model you pick. you can have the smartest model in the world, but if your routing is wrong, you're just burning money while your agents argue with each other in channels they shouldn't be in.

get the plumbing right first. the intelligence follows.

---

related reading

• 25 Things I Learned Using Claude Code Every Day
• Nature vs Nurture, But for AI Agents
• Your Meeting Notes Are Useless Until You Do This

---

this was originally published as an X article. subscribe to the early signal for more on AI infrastructure and agent setups.

Frequently Asked Questions

Can I run multiple AI agents through one Discord bot?

No. Each agent needs its own bot token. Sharing tokens causes agents to respond as each other and read from the wrong context. Create separate applications at discord.com/developers/applications.

How much does it cost to run 4 AI agents on Discord?

With proper channel isolation and routing, less than a junior employee's daily coffee budget per month. Without isolation, costs spike 4x because every agent processes every message.

How do I prevent AI agents from responding to every message?

Set requireMention to true on specialist agents so they only activate when tagged. Use allowlist group policy instead of open mode, and restrict each agent to only the channels relevant to its job.