pigeon: FAQ
Disclaimer
Some answers are humorous. If they mislead you, find another bot to use, because I don't want to serve you.
In a nutshell
pigeon is a minimalist AI chatbot for Discord. It is designed to listen to a singular channel and respond to all messages (excl bots) within it. These messages are stored in a chat history of up to 200 messages. pigeon can NOT process/see attachments, heavily search the web, or do any fancy stuff. It is intentionally lacking modern features. pigeon can also be called anywhere within a server @pigeon. These calls do not process external context (as in the past 5 messages), and these calls also do not contribute to your chat history context. pigeon can be used as normal in direct messages, but not everything is supported. Your chat can be exported with /history, cleared using /clearhistory, and imported using /import. Import files can be generated using /history, clicking "View raw data" and saving the page, such as CTRL+S or then on mobile.
Q: What is this?
A: pigeon is a simple AI Discord chatbot. You can talk to it in a server, or DMs. Just here
for a chat, you heard?
Q: I found a bug! How do I report it?
A: Email me! me@exerinity.com
If that bug also threw some errors, give me the timestamp of when it happened! Thank you in advance, Batman.
Q: Can I set custom instructions?
A: soon™
Q: What is the timer for?
A: pigeon employs a small timer for all messages for a few reasons:
- Debugging
- Used in tandem with the 2.5 second rate limit (read below)
- Aesthetical reasons. It's not in your way.
| Seconds | Label | How / What |
|---|---|---|
| < 2 | fast | casual chats or greetings |
| 2 – 9 | average | asking longer questions |
| > 9 | slow | writing large amounts of text, translating, or other complex tasks |
Q: When was it made?
A: September 2024. Idea coined July.
Q: Why is the model so stupid?
A: Because I'm still tuning the system instructions and constantly comparing models. If you
want top-tier professional always-correct answers, you're in the wrong place.
Q: Why won't it respond?
A: Check that the bot has permission to send messages in the channel. If it does, check the
set channel using /chat. If none is set, set one using the same command.
If it does not respond in direct messages, make sure you have DMs enabled from server members. If you do,
check the
mutual servers - one may have DMs temporarily paused. If it still doesn't work, email me.
Q: Are there limits?
A: In terms of per chat? Yes, 200 messages. I plan on raising this as development
progresses, but it's not important. In terms of a rate limit? Not that I'm aware of. I didn't set any hard
limits, but it may hit rate limits.
If you get over 200 messages, the oldest get overwritten, so it won't just suddenly "stop" when you reach
it.
There is a 2.5 second message cooldown per person to prevent spam. This is a generous limit -
Discord is very easy to spam quickly, but 2.5 seconds gives you time to respond.
Very long messages
are paginated into several messages.
Q: Do you store chats?
A: No, unless you run /history, which it is stored for 3 days. I
do not check them unless someone has notified me of misuse or bugs. "External server" is sugarcoating that
its a Cloudflare R2 bucket.
Q: I don't know how to use this! How?
A: RTFM
Q: Is this just a ChatGPT wrapper?
A:
Q: What model do you use?
A: Gemini 2.5 Flash. pigeon started off on GPT-3.5, moved to Gemma (Gemini's little
brother), then LLaMA 4 Scout. Gemini seems pretty sufficient but will most likely change again the future;
since pigeon is entirely experimental.
Q: Is it Cloudflare AI?
A: formerly
Q: Why is it called pigeon?
A: Because historically
pigeons delivered messages. The name was chosen arbitrarily - it was just a random word and I later
found a good reason for it. You know, because pigeon literally delivers AI generated messages?
Q: Can it process and/or generate images?
A: no, and never.
Q: How much does it cost to run this?
A: Basically nothing. Hosting is cheap and tokens are too. Roughly, under $15 a month.
Q: Can I discuss explicit content with it?
A: Probably not. I didn't tell it "no", but the model itself might refuse. After all, it is
Google.
Q: Why does the model change so much?
A: It doesn't. The last model switch was in July 2025.
Q: What is this error?
A: All errors are logged. If you see one, tell me when it happened, and I'll
investigate.
Upon errors, pigeon will retry your prompt 3 times. This should still work, but if it doesn't, send
it my way.
Common errors include "cannot read property 0 of undefined", which basically means
nothing; just re-send your message. If you get a long error about Google AI, it's self-explanatory from
there.
Q: Why is it so slow?
A: Because it isn't tuned for speed. Also, pigeon does not stream ongoing responses, it
only responds once the engine is finished; with everything it generated. Streaming is a bit of an illusion
to make it feel faster, but the times are roughly the same as GPT 4o.
Q: How did you make this?
A: Queue some big-room EDM and read a lot of documentations. It uses discord.js,
@google/genai to actually talk, and lots of if elses.
Out of all seriousness, the source code is available here: https://github.com/EX3Corporation/pigeon