Skip to content

AI TOOLS / MARCH – APRIL 2026

Discord AI Bridge

A Discord bot that answers questions with a model running on my own machine.

Built on my own.

Source code
MY ROLE
Solo project
KEY RESULT
Local inference moved off the event loop
DECISION
Move the blocking request off the event loop before considering a full async-client rewrite.
EVIDENCE
Source excerpt and an engineering note that explains the failure mode and trade-off.

How a request flows

  1. Discord!.ask command
  2. discord.pyEvent loop stays free
  3. Worker threadasyncio.to_thread()
  4. LM StudioQwen 3.5 9B, local

Why

I wanted to see what it takes to put a locally hosted language model behind a real chat interface, with no cloud API in between.

What I built

  1. 01

    Connected discord.py to LM Studio’s OpenAI-compatible endpoint and a local Qwen 3.5 9B model, triggered by the !.ask command.

  2. 02

    Moved the blocking HTTP call into a worker thread with asyncio.to_thread(), so the bot keeps handling other messages while a reply is generated.

  3. 03

    Turned connection failures, HTTP errors, and unexpected responses into plain messages in the channel instead of silent failures.

The code

discord_ai_bridge/main.py
View on GitHub 
try:
    # Run the blocking request in a thread so the bot stays responsive
    response = await asyncio.to_thread(requests.post, API_URL, json=payload)
    response.raise_for_status()
    ai_response = response.json()["choices"][0]["message"]["content"]

    await message.channel.send(
        ai_response, allowed_mentions=discord.AllowedMentions(everyone=False)
    )
except requests.exceptions.ConnectionError:
    await message.channel.send("⚠️ Error: LM Studio server is not running.")
except requests.exceptions.HTTPError as e:
    await message.channel.send(f"⚠️ Server Error: {e.response.status_code}")
The blocking call runs in a worker thread, so the event loop keeps serving other messages.

Outcome

People in the server can ask the local model a question with one command, and the bot stays responsive while it answers.

The trade-off

I kept the requests library and wrapped it in asyncio.to_thread() instead of switching to an async HTTP client like aiohttp. It fixed the blocking with one line and code I already understood. The cost is a thread per request and no way to cancel one once it starts.

What I learned

Async code only stays async if every slow call respects it. One blocking request can freeze the whole bot.

What I'd fix next

Add a timeout to the model request (today a hung model server holds a thread forever), and split replies longer than Discord’s 2,000-character message limit.

Go deeper

How one blocking call froze my Discord bot4 min read

Built with

  • Python
  • discord.py
  • LM Studio
  • REST API
  • asyncio

Working on something like Discord AI Bridge?

Tell me what you're building and I'll reply within two business days.

Open to co-op, internship and software roles in Vancouver, on-site or hybrid. Graduating May 2027.

Get in touch View resume All projects