slowpoke-typescript

slowpoke is a Discord bot converted from Rust to TypeScript, named after the Pokémon Slowpoke.

Features

• Chat and Prompt: GPT-5.6 text generation through an OpenAI-compatible API and LangChainJS
• Web Tools: Anonymous Parallel Search MCP tools for current web information and source extraction
• Image Generation: GPT Image 2 text-to-image generation through an OpenAI-compatible API
• Image Remix: GPT Image 2 edits of Discord-hosted PNG, JPEG, and WebP images
• Utilities: Ping, 8-ball, dice rolling, and TFTI commands

The deployment is intentionally fixed to:

• OpenAI-compatible API: https://codex.jeph.io/v1
• text model: gpt-5.6-terra
• image model: gpt-image-2
• Parallel Search MCP: https://search.parallel.ai/mcp

Setup

Local development requires Node.js 26 and pnpm 11.11.0. TypeScript remains on 6.0.3 because the current typescript-eslint release supports TypeScript versions below 6.1; TypeScript 7 currently crashes its parser.

1. Install dependencies:

   pnpm install

2. Create a .env file based on .env.example:

   cp .env.example .env

3. Configure the required runtime secrets:

   • DISCORD_TOKEN: Discord bot token
   • DISCORD_APPLICATION_ID: Discord application ID
   • CODEX_LB_API_KEY: API key allowed to use gpt-5.6-terra and gpt-image-2 through the configured OpenAI-compatible endpoint

The API key is sent only to https://codex.jeph.io/v1. The existing environment-variable name is retained for deployment compatibility. Do not commit .env or put runtime secrets in a Docker image.

Parallel's Search MCP is used anonymously and does not require an API key. Anonymous access has lower rate limits than authenticated Parallel plans. If the MCP cannot be reached during startup, slowpoke starts without web tools; text and image features remain available until the next restart.

Development

Run offline tests:

pnpm test
pnpm typecheck

Run the bot in development mode:

pnpm dev

Build and run in production:

pnpm build
pnpm start

Live provider smoke tests are manual and never run in CI. They use the local .env; image tests can incur provider charges and require a second explicit guard:

RUN_LIVE_AI_SMOKE_TESTS=1 pnpm smoke:live
RUN_LIVE_AI_SMOKE_TESTS=1 RUN_LIVE_IMAGE_SMOKE_TESTS=1 pnpm smoke:live -- --images

Docker

Use runtime environment injection rather than build arguments or image-level secrets. See docker-compose.example.yml for a minimal example:

cp docker-compose.example.yml docker-compose.yml
docker compose up -d

Commands

Slash commands

• /ping - Test the bot's latency
• /8ball [question] - Ask the magic 8-ball
• /prompt <text> - Ask the AI a question, with web search when available and useful
• /chat - Have slowpoke participate in the current conversation
• /tfti - Thanks for the invite
• /imagine <prompt> - Generate an image from text
• /roll [sides] - Roll a die

Prefix commands

• !remix <instructions> - Reply to a message containing a Discord-hosted or Discord-proxied PNG, JPEG, or WebP image and edit it

Remix downloads are limited to 10 MiB and do not fetch arbitrary external embed URLs.

Operational notes

• The current OpenAI-compatible gateway accepts gpt-image-2 generation and edit requests even though that image model is not advertised by GET /v1/models.
• GPT-5.6 dashboard cost accounting in gateway version 1.20.1 may be inaccurate; verify provider usage before relying on dashboard totals for budgeting.
• Anonymous Parallel MCP access is best-effort. A startup discovery failure disables web tools until slowpoke restarts.

Original project

This TypeScript version is converted from the original Rust implementation while adapting to TypeScript, Node.js, and Discord.js conventions.