CelestialBeing

AI Discord companion that answers /ask prompts with LLM responses, deployed as a serverless Discord HTTP interaction handler.

• Python
• Flask
• Discord Interactions
• Groq
• PyNaCl
• Vercel

A conventional Discord bot holds a WebSocket gateway connection, which means a process has to stay running. For a study companion that only needs to answer /ask, that operational cost is the wrong shape.

CelestialBeing handles Discord HTTP interactions on a Vercel serverless endpoint. A slash command hits the endpoint, the request is signature-checked, the prompt is sent to Groq with a short model fallback list, and the reply is returned. The function then idles.

• Flask handler under /api, deployed with vercel.json.
• Ed25519 signature verification of inbound Discord requests using PyNaCl.
• Groq inference with a documented fallback rotation across available models when a route is rate-limited.
• Slash command interface: /ask with a prompt, in channels or DMs where the app is installed.

HTTP interactions instead of a gateway bot

The workload is request/response. Serverless matches that better than a persistent socket, at the cost of in-process memory.

Verify before inference

Unauthorized HTTP calls should fail on the Discord signature, not on a missing API key after work has already been done.

Fallback models for availability

A single Groq model can 429. Rotating through a small list is a reliability tactic, not a quality claim about any one model.

Evaluation data has not yet been benchmarked. No published answer-quality scores, latency distributions, or uptime measurements beyond the architecture’s intent.

• Stateless by design: there is no documented long-term memory across turns.
• Capacity is bounded by Groq and Vercel free-tier limits; “24/7” here means the endpoint can sleep and wake, not a measured SLA.
• Study/Q&A companion — not a tool-using agent.
• A Vercel URL is listed on the repository; treat it as a deployment pointer, not as evidence of production usage.