Give your support channel a break.
Save a moderator’s reply.
ThreadSherpa uses it to answer similar questions in your Discord forum.
People miss
earlier answers.
A moderator has already answered. The next person starts another thread and asks the same question.
Where do I submit my project?
Is a GitHub link enough?
Where should I send my project?
Save an answer.
Right-click a moderator’s reply, then choose Apps → Save this answer.
The bot reuses only the replies a moderator saves.
The bot watches
the forum.
ThreadSherpa checks new forum posts against saved answers. When one matches, it writes a reply with a link to the original.
Is a GitHub link enough to submit?
It reads earlier replies before answering a follow-up.
Questions it
can’t answer.
When no saved answer covers a question, the bot stays quiet and queues it for a moderator.
Architecture.
How the bot checks each new forum post.
- Discord
Read the question
The forum title, post and replies stay in one conversation.
- MongoDB
Find a saved answer
Exact match, then text and local embeddings. Scoped to this server.
- Paper / Ollama
Write a short reply
The model gets the question, context and approved evidence.
- Response policy
Check the source
Check citations and links before posting. Unsupported answers wait.
No matching saved answer?
Skip the model. Queue it for a moderator.
What gets stored?
The approved answer, its original question, source and scope. Duplicate saves reuse the same knowledge. A moderator chooses any optional web sources. Read the memory code.
Limitations.
It does not learn from every message.
Only replies saved by a moderator are reused.
When the rules change, update the saved answer.
Edit or remove the saved guidance. Future questions use the current facts. The bot does not broadcast corrections into old threads.
The model can still get things wrong.
Replies include their source so a person can check them. Unsupported questions wait. An account or payment issue needs private human support.
You run the bot and choose the model.
Use local Ollama inference, or Paper Instruments for replies with local embeddings. You also choose the database, support channels, and moderators.
I built this after seeing the same support questions in the MLH Discord.
Different people needed the same information, but the useful answer was buried in another thread.
Made for the Hacktoberfest Open-Source AI Challenge: Week 3 on DEV. The theme is Slow and Steady: build something that does the work so you don't have to. This is a hackathon project, so expect rough edges.
- Create your Discord application.
- Add the environment values and start the bot.
- Connect a forum. Save the first useful reply.